Embed Size (px)
Citation preview
A Feature-based Cost Estimation Model
by
Narges Sajadfar
A thesis submitted in partial fulfillment of the requirements for the degree of
Master of Science
in
ENGINEERING MANAGEMENT
Department of MECHANICAL ENGINEERING
University of Alberta
© Narges Sajadfar, 2014
ii
Abstract
To address the requirement of dynamic pricing and cost control in high-variation
product manufacturing, nowadays many companies face the problem of generating
quotes and order prices timely, accurately and consistently. This thesis reports a
preliminary investigation on automatic cost estimation with a feature-based semantic
model.
A generic semantic model for the purpose of automatic cost estimation is proposed, in
which a new concept of cost feature, is suggested. A cost feature can be identified with
data mining methods for different clients or products, and conceptually interfaced with
product design and manufacturing features. Feature-based mapping models can be used
to determine feature scope and cost level definition, including all the dependency
relationships with other domain features. This model is expected to enable a visual,
flexible and semantically consistent scheme to address effective and efficient product
cost structures, frequent configuration variations and business changes.
Cost feature has been defined in this work as a unique class in the unified feature
modeling system to address the characteristics of cost engineering entities, constraints
and dependency relations. This research also describes the relationship between a cost
feature and three engineering sub-models, i.e. machining model, design model and other
auxiliary data model by associating tangible and intangible data. Further, semantic
relations are investigated in early product design process for dynamic, accurate and
iii
visible product cost estimation. This research also discusses cost engineering related
functions, associated data structures, and techniques to be used.
Further, this research presents a Cost Estimation (CE) method that is tailored to apply
feature-based engineering concept with data mining algorithms. The method proposed
combines and leverages the unique strengths of linear regression and data mining
approaches, and is based on a mechanism to discover cost features. The final estimation
function takes the user’s confidence levels up by phasing in the real time application of
the method gradually and building up the data mining capability.
A case study is presented to demonstrate the hybrid method, and it compares the
results of empirical cost prediction via data mining. The case study results indicate that
the combined method is powerful and accurate for determining the costs of the illustrated
welding features presented. With the results of the empirical prediction of five different
data mining algorithms, the most accurate algorithm for welding operations is identified
and presented in more detail.
iv
Preface
This thesis is an original work by Narges Sajadfar. The semantic modeling approach
that used in chapter 3 and 4 was published in two book chapters:
-Sajadfar, Narges, Yanan Xie, Hongyi Liu, and Y-S. Ma. "Introduction to Engineering
Informatics." In Semantic Modeling and Interoperability in Product and Process
Engineering, pp. 1-29. Springer London, 2013.
-Sajadfar, Narges, Yanan Xie, Hongyi Liu, and Y-S. Ma. "A Review of Data
Representation of Product and Process Models." In Semantic Modeling and
Interoperability in Product and Process Engineering, pp. 31-51. Springer London, 2013.
A version of chapter 3 has been published in Journal of Engineering and
Technology. Ma, Y. S., Narges. Sajadfar, and Luis. Campos Triana. "A Feature-Based
Semantic Model for Automatic Product Cost Estimation." IACSIT International Journal
of Engineering and Technology 6, no. 2 (2014).
A version of chapter 4 has been published in International Journal of Mechanical
Engineering and Mechatronics. Sajadfar, Narges, Luis Campos Triana, and Ma, Y. S..
"Interdisciplinary Semantic Interactions within a Unified Feature Model for Product Cost
Estimation."
A version of chapters 5 and 6 has been submitted to Advanced Engineering
Informatics. Sajadfar, Narges, and Yongsheng Ma. (Revision)
Technical case study and data collection were provided by McCoy Drilling and
Completions (Farr) Company.
v
Acknowledgement
I would like to express my greatest thanks and appreciations to my supervisor, Prof.
Yongsheng Ma; for his guidance, help, and support. I am very grateful for his invaluable
contributions to my personal and professional development. I could not have asked for a
better supervisor.
I express my deepest thanks to McCoy Drilling and Completions (Farr) Company, for
giving me the opportunity to complete MITACS Accelerate Cluster internship program
for two years. I would like to thank McCoy supervisors: Jose Pazuengo, Julio Barradas
and Luis Campos for their tremendous support. I would like to extend my appreciation to
my coworker at McCoy: Yanan, Tony, Henry, Parthiban, and Pedro. They created an
enjoyable working environment.
I am so grateful to all my lab mates and friends in MECE 4-23 lab for creating a warm
atmosphere: Fatemeh, Qingqing, Lei Li and Moin.
My special thanks go to my dear friend Afsaneh Esteki who helped me to learn and
work with data mining software by sharing her knowledgeable advice.
I would also like to take this opportunity to thank colleagues and friends, Nahid,
Fatemeh, Fahimeh, Neda, Leila, Parisa, Sara, Roya, and Negar for their support, care and
unforgettable friendship.
And finally but importantly, I am also thankful to my parents and my sisters for all
their love and unconditional supports throughout my life. Also, my sincere appreciation is
given to my husband Mohammad Mahmoudi, for his encouragement and continues
support.
vi
Table of Contents
Chapter 1: Introduction ................................................................................................................... 1
1.1 Background .......................................................................................................................... 1
1.2 Problem statement ............................................................................................................... 2
1.3 Research challenge .............................................................................................................. 4
1.4 Research goals ...................................................................................................................... 5
1.5 Research scope ..................................................................................................................... 6
1.6 Research methodology ........................................................................................................ 8
1.7 Thesis organization .............................................................................................................. 8
Chapter 2: Literature review ......................................................................................................... 10
2.1 The need for cost engineering ........................................................................................... 10
2.2 Cost types ........................................................................................................................... 11
2.3 Classification of product cost estimation techniques ...................................................... 13
2.3.1 Qualitative techniques ................................................................................................ 14
2.3.2 Quantitative techniques ............................................................................................. 16
2.4 Review on feature-based cost estimation ......................................................................... 18
2.5 Review on software tools for cost estimation .................................................................. 21
Chapter 3: A feature-based semantic model for hybrid production cost estimation ................... 23
3.1 Cost estimation model ....................................................................................................... 23
3.2 Cost feature and cost feature association ........................................................................ 26
3.3 Semantic modeling ............................................................................................................ 28
3.4 Feature-based semantic cost modeling ............................................................................ 31
3.5 Conclusion .......................................................................................................................... 36
Chapter 4: Semantic Interactions within a Unified Feature Model for Product Cost Estimation . 37
4.1 Design model ...................................................................................................................... 37
4.2 Manufacture model ........................................................................................................... 41
4.3 Auxiliary data model ......................................................................................................... 43
4.4 Data obtained by employed sub-model ............................................................................ 45
4.5 Data mining for cost feature pattern construction ......................................................... 46
4.6 System design for cost estimation .................................................................................... 49
vii
4.7 Integration among sub-models ......................................................................................... 50
4.8 Conclusion .......................................................................................................................... 54
Chapter 5: A Hybrid Cost Estimation Method Based on Feature-oriented Data Mining ............. 55
5.1 Current practice of cost calculation in ERP systems ..................................................... 55
5.2 ERP solutions ..................................................................................................................... 57
5.3 The principle of hybrid cost estimation method ............................................................. 59
5.4 Definition of the cost feature ............................................................................................ 62
5.5 Empirical cost estimation process .................................................................................... 66
5.6 DM cost estimation process .............................................................................................. 69
5.7 Data mining implementation in the cost engine .............................................................. 72
Chapter 6: Case study – welding process cost estimation ............................................................ 73
6.1 Definition of welding process features ............................................................................. 73
6.2 The data structure of welding feature ............................................................................. 76
6.3 The data structure of welding cost feature ...................................................................... 79
6.4 Mapping of welding manufacture feature and the associative cost feature ................. 80
6.5 Prototype results and discussion ...................................................................................... 81
6.6 Conclusion .......................................................................................................................... 88
Chapter 7: Conclusion and future work ........................................................................................ 90
7.1 Conclusion .......................................................................................................................... 90
7.2 Future work ....................................................................................................................... 91
References ..................................................................................................................................... 92
viii
List of Tables
Table 2.1 Total product costs category .......................................................................................... 13
Table 3.1 ERP information about safety door part ........................................................................ 33
Table 3.2 Cost feature information ................................................................................................ 35
Table 4.1 Detail design information for work pieces before welding ........................................... 41
Table 4.2 Manufacturing Information for the door weldment. ...................................................... 43
Table 4.3 Generated output by integrated model. ......................................................................... 48
Table 5.1 Cost feature template ..................................................................................................... 65
Table 5.2 Regression Statistics Result ........................................................................................... 68
Table 6.1 Data structure of welding feature properties ................................................................. 78
Table 6.2 Sample CE based on regression analysis....................................................................... 82
Table 6.3 correct matching and error rates comparison between five different algorithms to the
historical data set (%) .................................................................................................................... 85
ix
List of Figures
Figure 2.1 Cost commitment curve ............................................................................................... 11
Figure 2.2 Fixed and variable cost ................................................................................................ 12
Figure 2.3 Initial classification of product cost estimation ........................................................... 14
Figure 3.1 Cost feature related elementary semantics with some non-exhaustive relations ......... 25
Figure 3.2 Concept of semantic processing model for the proposed cost estimation .................... 27
Figure 3.3 Example Tong Assembly Body 3D CAD model ......................................................... 31
Figure 3.4 Safety door model ........................................................................................................ 32
Figure 3.5 Cost feature elements for the safety door module ........................................................ 34
Figure 4.1 Cost feature semantics with non-exhaustive relations ................................................. 39
Figure 4.2 Assembly drawing of door weldment and its individual work piece ........................... 40
Figure 4.3 Auxiliary cost modeling with data mining process. ..................................................... 48
Figure 4.4 Computer system diagram for cost estimation ............................................................. 52
Figure 4.5 UML diagram of integration between sub-model features. ......................................... 53
Figure 4.6 Difference between machining features and geometry design features ....................... 53
Figure 5.1 Typically-defined cost attributes in EPICORTM
........................................................... 58
Figure 5.2 An example set of cost attributes defined in Visual InfoTM
......................................... 59
Figure 5.3 A hybrid cost estimation method proposed .................................................................. 61
Figure 5.4 Cost feature definition in UML format ........................................................................ 63
Figure 5.5 Relation between total cost and amount of welding wire ............................................ 67
Figure 5.6 Empirical cost estimation process ................................................................................ 68
Figure 5.7 Sample of scatter plot of the multiple regression analysis ........................................... 69
Figure 5.8 DM cost estimation configuration ................................................................................ 71
Figure 6.1 A typical welding feature ............................................................................................. 73
Figure 6.2 Welding feature and cost feature structure relations .................................................... 75
Figure 6.3 Data structure of Welding feature class ....................................................................... 76
Figure 6.4 Geometric Parameters of welding in different welding form ...................................... 77
Figure 6.5 Data structure of welding cost feature class ................................................................. 79
Figure 6.6 Sample result of CE based on the regression analysis of weka ................................... 84
Figure 6.7 User Interface (UI) for welding cost estimation .......................................................... 87
1
Chapter 1: Introduction
1.1 Background
In today’s competitive world and the on-line business era, effective and timely cost
estimation has a critical role in manufactory’s success. In terms of practical applications,
industry needs to quickly estimate the price for any new product configuration in a
customer order either interactively or automatically by on-line applications. High cost
estimation accuracy at the very early stage and effective cost control during the operation
phase are essential for the smooth business operation. Quick and accurate cost estimation
brings lots of advantages over competitors. Cost estimation can also affect all aspects of
production processes and management decisions, such as choosing between producing
the product components in-house or outsourcing them.
The common challenge of most companies is to improve profitability. Historically,
timely and accurate cost estimation has an important role in a manufacture’s success that
can improve the manufacturing performance and effectiveness of business [1].
Companies follow different strategies and practices that can increase the revenue and
decrease expenses. Cost engineering research is imperative to guide companies by
developing tools and models for effective and efficient cost estimation and cost control.
Cost estimation is an important process for management decision making. For
example, the manufacturing manager needs to provide detailed and accurate cost
estimation for making the strategic decisions, such as determining the bid price for
product, make-or-buy decisions, marketing approach and setting the goals of profitability.
To achieve all the benefits of accurate cost estimation, the manager needs to formulate
2
the best suitable cost estimation method that best matches the specific products’
structures and individual business nature.
In addition, cost estimation can be carried out in each stage of a product life cycle,
such as conceptualization, concept design, detailed design, prototyping and manufacture.
Many researches recognize that up to 80% of the product cost is determined at the
conceptual and design phases [2]. Therefore, accurate cost estimation at early stages with
different levels of defined details has an important role in manufacture success.
1.2 Problem statement
In manufacturing, accurate cost estimation and tracking cost estimation are well-
known problems, and a definite solution to these is nonexistent. The traditional cost
estimation approaches focused on parametric, analogy and bottom-up techniques and
they are still being used in many companies of different industries [3]. Most financial
managers rely on empirical cost estimation as a decision tool. However, in today’s
competitive world, with a large variety of business models, products and purposes, cost
estimation should be done in a comprehensive system that can include all the details, and
calculate the cost automatically and accurately.
In the past two decades, Enterprise Resource Planning (ERP) systems have been
essential contributors to the success of costing advancement in manufacturing companies.
The development and implementation of ERP systems helps industries to gather and
manage all their information into a common computer-based information system [4]. By
utilizing integrated ERP systems in manufacturing, it seems true that most of the cost
calculation issues were addressed [5]. The ERP cost calculation can be typically for a
specific segment in a production cycle, for specific machining and manufacturing
process, or for a specific part and component. ERP cost calculations are based on
parametric cost estimation that requires all data for the details of parts or processes, but
3
such complete data is usually not available, particularly in the dynamic design and early
sales states where prototypes are being built and configurations continuously modified.
In addition, manual variations of processes, such as welding, are not well defined and
documented on ERP systems; and companies with small batch size and high variation
production have difficulties to model product costs accurately.
This research addresses cost data availability and verification issues by proposing a
new cost estimation modeling approach based on a semantic feature-based cost
estimation engine, together with the use of a cost feature library and system integration.
However, such implemented methods are currently limited to simple cost estimation, and
depend on approximations and assumptions. Even the most complete ERP system has
numerous limitations regarding the cost estimation. For instances:
- Specific organization or manufacturer cost scenarios are not considered in ERP
costing data setup [6]
- All cost-details data have to be provided for accurate cost calculation which
requires a lot of time, and the information used may not be accurately assessed
- Missing data has a huge negative effect on cost estimation accuracy and
calculation
- The existing ERP systems could not calculate the cost base on products
configuration and machining features
- There is not any reusable process for gathering information and cost calculation
- The historical data is often not considered in ERP cost calculation
- The particular company costing strategy is not effectively reflected in ERP cost
implementation
- The calculation of a new product cost is not accurate due to lack of information
4
1.3 Research challenge
Manufacture cost estimation is a cluster of challenging tasks done in three levels:
conceptual design, detail design and manufacture. Although cost estimation is clearly
related to the profitability, the systematic pricing was not adopted in most companies due
to the complex associativity among the technical and business considerations. A
systematic semantic model is required in order to tackle the relations and constraints
among product elements, processes, resources and actors.
Many manufacturers used to do their cost estimation at the manufacturing process
level; this approach represents the traditional way, and is also known as engineering
bottom-up method [7]. In the manufacturing process level, all the expenses can be
quantified already. For example, Donald R. Woods mentioned more than 20 factors in his
book, such as: cost of fabricated equipment, cost of material, cost of labor, cost of
control, cost of land, cost of storage, legal fees [8], etc. Although this traditional approach
is still being used, however its drawback is pretty clear, that is when the product gets
complex, the accuracy of cost estimation will decrease significantly. Also there are lots of
uncertain factors behind all the cost elements, such as the changes of costs related to the
changes of materials, processes and equipment items. Such uncertain dynamic changes
can increase the risk of broad range of cost estimation, such as the risk of misleading
loss-indication in a potentially profitable deal [9], or the opposite.
Cost estimation is a dynamic procedure that must be done periodically during each
stage of a product lifecycle; and it has been the trend to compete on price at product
conceptual specification and the early design level. So instant and yet more accurate cost
estimation is demanded at the early stage of product development. However due to the
lack of accurate information in the early stage of a product, or a business order, the cost
estimation is currently carried out with experience and subjects to significant
modifications at the manufacture level.
5
The purpose of this research is to introduce a new semantic model that can
systematically associate design features and manufacturing features with a cost
estimation mechanism, i.e. the proposed cost features. They can serve as the active
information agents and work out cost elements intelligently in the product specification
stage with the inference capability from the historical and expected cost models of
associated features at the design and manufacture levels.
1.4 Research goals
The top objective of this research is to improve the accuracy of production and
machining cost estimation by defining a set of cost features and developing a prototype of
a cost estimation engine. This thesis also reports the implementation of new feature-based
cost estimation method by using the cost feature concept and a hybrid data mining and
empirical algorithm. With the proposed approach, cost estimation can be partially
automated with a gradual adoption process with the user-controlled. The feature-based
cost estimation method involves the following steps:
- Identifying ERP cost estimation limitations
- Develop a method calculating manufacture costs by identifying the processes,
involved machines and tools for each product component and manufacturing
processes
- As a test case defining cost estimation scenarios with newly defined welding
features
- Gathering all the cost-related information and deposit the data into a database
- Developing a computer-based cost simulation tool and using it to estimate product
cost scenarios
6
1.5 Research scope
This research, presents a novel feature-based cost estimation method by combining
two advanced cost estimation approaches, empirical and Data Mining (DM) by using cost
feature concept. A prototype software toolkit is also implemented. Cost feature is defined
as a set of characteristic attributes of an object that can be associated and constrained to
represent the patterns of semantics for the purpose of cost engineering. The automation of
cost estimation based on cost features includes three steps:
- Gathering historical data:
Our selected company’s ERP has the information about sales, inventory,
purchasing, and manufacturing of products. Based on project needs, we can
transfer information from the company’s ERP into our new database, where we
have more control on the existing data and we can do the necessary analyses.
- Abstraction of cost feature and cost feature data extraction:
Each product has different characteristic engineering attributes such as: design
features, functions, suppliers, tooling, machine profiles, machining features, etc.
and each of this attributes have their own influence on cost. Ideally, the cost
estimation feature could make predictions based on product functions (f) and their
weights (W) for each manufactured product. For example if a product includes
components A, B, and C. The cost feature for this product is:
C total = WA * f (A) + WB * f (B) +WC * f (C) (1-1)
Abstraction of feature related to costs should be generically workable for all
different parts, modules, sub-assemblies and products that is describe in chapter 5
and 6 with more details.
The database that we have developed for this project contains information for Bill
of Materials (BOM) and Method of Manufactures (MOM) of 500 different
manufactured products from the company partner. Additionally, the cost engine
7
has to recognize and extract cost feature data in order to obtain the cost
estimation.
- Using different algorithms and methods to find the most accurate cost estimation
Based on data mining (DM) techniques, the cost engine has to select one effective
cost estimation algorithm out of several available options to gain a trustable cost
estimation result with measurable confidence level. WEKA is a machine learning
software tool that provides a comprehensive collection of algorithms for
automatic classification, regression, clustering, and feature selection [10]. WEKA
has been used in this research as a tool in order to support the cost feature
identification and extraction, and application. The DM results provide a guide for
selecting the most accurate algorithm based on the available data sets.
To complete the evaluation cycle for the proposed approach, the data set collected
from ERP historical records is divided into three subsets, stored in separated databases:
training data, testing data, and application data. WEKA uses the training data to generate
specific algorithms. Then WEKA uses this algorithm on the testing data to estimate the
costs. With the testing database, the actual costs are also known, and then the reported
costs can be compared with the results from the proposed cost estimation algorithm. If
the accuracy of cost estimation is acceptable, then WEKA algorithm is used to estimate
the costs of the products in the application database, then the final effectiveness is
measured by a confidence level for the feature application of the cost estimation engine.
8
1.6 Research methodology
Feature technology has been successfully applied in engineering computer tools in the
past decades because of its capability to resemble engineer’s semantic patterns related to
their design and manufacturing threads. Feature is a kind of well-defined and dynamic
data structures that can describe more information related to all the aspects of the product
that is not reality described in traditional analytical models or semantic attributes [11].
Many advanced engineering design features and machining features have been
successfully defined and applied. Features can describe the material characteristics,
machining methods, geometric shapes, etc.
The feature-based cost estimation approach entails to identify cost related features that
can be defined as associated to certain functions and manufacturing processes. For
example, a part’s cost can be linked to the design features for the estimation of material
cost, power, optional add-ons, and other quantifiable impacts. Cost can also be linked to
manufacture features to estimate the machine utilization and process time cost [12].
However, so far in practice, identifying, defining and the tracking the dependencies and
impacts of changes related to design and machining features at each level of cost
estimation are very complicated procedures and almost impossible for a complex product.
1.7 Thesis organization
Relevant methodologies and techniques in the areas of cost estimation and feature-
based modeling which are previously investigated by other researchers are first reviewed
in Chapter 2.
Chapter 3 discusses the concept of proposed approaches by introducing research
challenges and methodology. The goal of this chapter is to investigate a concept of new
manufacturing cost calculation model coherently throughout the lifecycle of a product
9
series, especially emphasizing at the conceptual design stage with the input of expected
manufacture process information. The proposed model integrates three functional sub-
models: feature-based costing, data mining, and semantic reasoning.
Chapter 4 is devoted to a more comprehensive description of the proposed system:
- A data structure that uses the unified feature concept to support data associations
and sharing across the design model, machining model, and auxiliary data model.
- Inherent relations among cost feature and other features, such as design features and
machining features.
The purpose of this chapter is describing a new cost estimation functional module and its
semantic interactions within the above-mentioned three sub-models and illustrating the
effective definition and use of the cost feature objects which can be directly usable in
computer functions for the cost estimation at different engineering phases or levels, such
as the design and manufacture stages.
Chapter 5 focuses on the discussion of tools and procedures used to carry out the
hybrid cost estimation method based on feature-oriented data mining:
- Definition of cost feature and its requirement
- Empirical and data mining cost estimation processes
- Implementation of data mining algorithms in the proposed cost engine
Research result and case study are discussed in Chapter 6 followed by conclusions and
future work recommendations.
10
Chapter 2: Literature review
2.1 The need for cost engineering
The American Association of Cost Engineers (AACE) describes cost engineering as
“that area of engineering practice where engineering judgment and experience are
utilized in the application of scientific principles and techniques to the problem of cost
estimation, cost control, and profitability” [13]. Cost engineering tools help companies
with decision-making, cost management and budgeting control [14]. Cost engineering
includes cost control, cost forecasting, risk analysis, investment evaluation and
profitability analysis. However, cost engineering focuses more on cost estimation and
cost control.
In addition, cost engineering during the early stage of product lifecycle has an
important role in product development. Many researchers agree that between 60% to 80%
of a product cost comes from the concept phase [2, 15, 16]. Figure 2.1 shows a typical
cost commitment curve during the concept and production phases. In average, decisions
made during concept design phase contribute up to 70% on the total cost of a finalized
product. However, the actual cost incurred in the conceptual design level is less than 20%
when customized products are developed, specification of configurations affect the total
cost significantly [17].
Also, many decisions need to be taken during the further stages of the product
development cycle such as: make-or-buy, manufacturing process plan, recycling of the
product, etc. The decisions are based on different criteria and cost estimation has an
important role to influence the decisions. However, limited amount of available data
11
during the early stage, uncertainly variables and unknown conditions make the cost
estimation accuracy questionable.
Figure 2.1 Cost commitment curve [17]
2.2 Cost types
The definition of manufacture costs is the sum of all costs of the resources (labor,
equipment, tools, materials, energy and etc.) that was used in the process of making a
product [18].
All the manufacturing costs can be divided into two main categories: fixed costs and
variable costs. Fixed costs are those cost that do not change based on a quantity and it’s
independent of output. On the other hand, variable costs are based on the products
quantity and it’s growing by increasing the output [19]. In addition, cost categories can be
broken into direct and indirect costs. Direct costs are directly and consistently connected
to the production of the product such as machining operations, service costs, software and
tool. Indirect cost cannot be recognized or pinpointed for specific products or parts such
12
as building rent, utilities, administration, maintenance, and staff salaries [20]. Figure 2.2
illustrates the relation among fixed and variable cost and quantity of products. Fixed cost
has a flat line while variable cost is increasing by number of products.
Figure 2.2 Fixed and variable cost
Different categorization methods for cost items can be formal in [21]. Table 2.1 shows
an example of total product cost categories which introduced by Asiedu, 1998 [22]. At
the first level, total cost is divided to four main categories: research & development cost,
production and construction cost, operations and maintenance cost and retirement and
disposal cost. Next level describes more details of each category. For instance:
- Product software belongs to research and development cost which is direct and
fixed cost.
- Quality control belongs to production and construction cost that is indirect and
variable cost.
Co
st $
Fixed cost
Variable cost
Total cost
Produced Units
13
Table 2.1 Total product costs category [22]
Total products cost
Research & development
cost
Production and
construction cost
Operations and
maintenance cost
Retirement and disposal
cost
- Product management
- Product planning
- Product research
- Design documentation
- Product software
- Product test and
evaluation
- Manufacturing/
Construction
management
- Industrial engineering
and operations analysis
- Manufacturing
- Construction
- Quality control
- Initial logistic support
- Operations/
maintenance
management
- Product operation
- Product distribution
- Product maintenance
- Inventory
- Operator and
maintenance training
- Technical data
- Product modification
- Disposal of non-
repairable
- Product retirement
- Documentation
2.3 Classification of product cost estimation techniques
There are various techniques and methodologies for product cost estimation with their
own advantages and disadvantages. This chapter contains a review of the product cost
estimation literature, and focuses on the techniques and software tools used. We can
classify cost estimation techniques (CETs) into different ways, such as: traditional and
modern ones in the sense of adoption time; qualitative and quantitative from the analysis
methods; top-down and bottom-up from reasoning approaches; and parametric and
judgmental cost estimation from data input and propagation angle.
Traditional cost estimation approach includes bottom-up engineering, analogy, and
parametric methods. They are fundamentally based on historical data and experience
knowledge. The advanced cost estimation approaches include expert judgment, feature-
base evaluation, fuzzy logic, neural network and data mining (DM) methods; they are
more suitable for complex products and dynamic cost estimation [23].
14
Regarding the accuracy of cost estimation adopted from [12], all the techniques can be
divided into two groups as figure 2.3 shows: qualitative and quantitative ones. Qualitative
cost estimation is used when it is difficult to accurately describe the cost based on the
product data, and it requires heuristic methods [24]. Usually, this technique requires
judgment of an individual with expertise to identify the similarity between the studied
product and previous products in order to relate their costs [12]. Quantitative cost
estimation is based on product details and processes data. Historical data and statistical
models have a main role in quantitative cost accuracy [25]. In the next level, qualitative
technique includes intuitive and analogical methods; and quantitative techniques include
parametric and analytical methods [12].
Figure 2.3 Initial classification of product cost estimation [12]
2.3.1 Qualitative techniques
Qualitative CE uses a heuristic method in the case of difficulty in finding the accurate
cost based on parametric product data [24]. Intuitive cost estimation is the first category
from qualitative CE which is a fast CET that use experienced individual for cost
estimation. However, each individual (expert) is likely to include his personal input and
thus is likely that two expert individuals will come up with different cost estimations;
Cost estimation
techniques
Intuitive
Qualitative Quantitative
Analogical Parametric Analytical
15
which will certainly decrease cost estimations accuracy. To solve this issue the researcher
has defined different systems and methodologies to create a more accurate framework for
this CET [26]. Intuitive techniques can be further largely subdivided into two approaches:
case-based reasoning and decision support system-based.
- Case-based reasoning (CBR) uses an expert person to find a past design similar to
the new design and associate their costs. The past design will be adjusted to become
similar to the new design, and then the cost of each change will be calculated [27].
A smarter way in this approach is using an information processing technique to
extract geometrical features from the product Computer-aided design (CAD)-
model. Then the similar cases are identified and the costs suggested by a system for
cost calculation [28]. Usually this technique is used through conceptual phase or
design phase to decrease the total cost of the product.
- Decision Support System (DSS) is a management tool, which can be defined as a
theoretical system to describe the processes of decision-making or the computer-
based system that is used to represent the information of a decision-making process
[29]. DSS can be used for cost estimation purpose and implemented based on three
techniques: rule-based system, expert system and fuzzy logic system. Each of these
techniques provides a framework for decision-making process in cost estimation
and needs to have manufacturing process knowledge and practical experience to
make it practically usable [30].
• Rule-based system: This is the most common technique for DSS, which is
an Artificial Intelligence (AI) application. This technique needs to identify a
pattern and structure from data to identify applicable situations, and then it
has to define an appreciate action for each situation. This AI technique was
introduced in 1960s, but this method was not used for cost estimation until
the end of 1980s [31]. Kingsman Brian et al, [30] identified around 200
heuristic cost estimation rules for make-to-order system. Gayretli, A. et al,
[32] introduced a rule-based system for feature-based cost estimation. They
extracted the form feature and process feature from CAD model, and then
16
the available processes are suggested by rule-based system based on feature
attributes process capability and production rules. In the same time, the rule-
base system will calculate the cost based on the programmed functions.
• Expert system is the first commercial artificial intelligence program that
simulates the human judgment by using expert knowledge. The purpose of
using expert system is to help general managers to make better or faster
decisions, or it can be used partially instead of a professional for basic
decision making in a specific area. Expert system contains two main parts:
interface engine and knowledge base module. Knowledge can be store based
on different facts and rules into knowledge-based module and interface
engine suggests the solution based on the pre-defined rules stored in the
system [33].
Analogy cost estimation is a qualitative technique that compares the new product with
the past product based on functions and geometrical similarity [25]. The degree of
similarity has a direct effect on accuracy of analogy cost estimation. Usually, past
products require changes to arrive the new product information which needs time and has
to be done by expert professionals. In addition, there is not a unified framework for
analogy cost estimation and each expert person can do it in different ways. Thus, their
results are not comparable. Analogical techniques are further subdivided into Regression
analysis models and Back propagation neural network (BPNN) models.
2.3.2 Quantitative techniques
The quantitative CE approach is focused on analytical cost calculation functions and
the exact parameters summarized from the records of business transactions for costing
purpose [12]. Quantitative techniques have two main categories: parametric and
analytical techniques.
Parametric cost estimation (PCE) is a principle method for cost estimation in early
stage of product life cycle. Usually there is not enough information about the product in
17
conceptual or design stage. PCE can provide a statistical relationship between variables
based on previous product. As a result, the build statistical formula can be used for new
product cost estimation [24]. In general, PCE tries to define the relation among product
characteristic and cost information. Then, it uses the extracted methods for top-down cost
calculation. Thus, it can be described as a parametric model that contains different
formulae [25]. Many techniques and methods can be used to extract the statistical
relationships. They can be categorized in three main methods: Statistical methodology,
linear regression, approximate tool paths and process parameters. Linear regression is the
most common technique using statistical methodology to measures the dependency of
selected variables in relation to other independent variables [34].
Analytical techniques are focusing on detailed design. The purpose of analytical
techniques divide the product based on their activities, operations and properties to create
more details and information regarding to the specific parts or products [35]. Feature-
based cost estimation and activity-based cost estimation are two common techniques in
analytical techniques categories.
- Activity based:
Activity based costing (ABC) is a costing approach that outputs the cost of products
based on the activity being performed on the product. ABC has two main central parts:
activities and cost drivers. The cost of each activity has to be calculated first, and then it
will be assigned to the product. The total cost will be calculated by adding each activity
assign cost [36]. This type of CE is very useful for manufacture cost estimation which
can calculate the cost based on the manufacturing activities. ABC is proposed by
reference [37]. Through investigation on traditional cost estimation, they suggested to use
ABC for calculating overhead cost instead of using fixed overhead cost.
18
- Feature-based:
Feature-based CE can be very much detailed and accurate for complex products; this
method basically decomposes product manufacturing processes into cost factors based on
their operation methods, features, surface quality requirement, key process dimension and
geometrical tolerances, and supporting activities,. In order to manage the tedious cost
elements more efficiently while keeping the accuracy of CE, the correlated cost patterns
can be identified and categorized with product configuration variations and predefined
procedures. In other words, analytical examination can be extended to a feature-based
approach.
2.4 Review on feature-based cost estimation
In this research, a feature-based CE approach is proposed as an analytical CE
technique to identify cost-related features that are associated with certain function
specification and manufacturing processes [12].
Feature recognition goes back to Kyprianou’s PhD work at Cambridge in the late
1970s. In 1980s, Shah et al. [38] introduced the geometric feature reorganization from
CAD model, they mapped feature concept with engineering semantics and software
engineering object-oriented approach conceptually. Later, Emmerik V et al. [39]
introduced a set of graphical user interfaces for feature-based modeling. The feature-
based engineering informatics approach has been gaining more and more popularity into
computer-aided software tools, such as in CAD and CAM domains. However, this model
has not been generally used for cost-determining purposes.
Leo Wierda [40, 41] is one of the first researchers who used the concept of feature in
qualitative and quantitative CE. In 1990, he introduced a new tool that would enable the
user to effectively control costs. He defined the “life cycle cost” by three phases in the
19
model, including conceptual design, materialization, and detailed design. He also defined
two types of cost information: design rules and manufacturability information. Design
rules refer to the historical design information used to describe a framework for a current
design. Manufacturability information refers to the current design information and
geometry information of a product; this information is used to extract the product cost
data. In his cost model, the target cost is considered in each phase. As a result, the design
is based on cost, and the cost is based on information derived from the feature
reorganization program [40]. In 1991, Leo Wierda expanded “life cycle cost” to include
feature-based cost modeling. He used the feature-based model to fill in the gap between
the design and the process plan. In addition, he described the advantages of calculating
the costs per features [41].
By the end of the 1990s, more researchers focused on feature-based CE, and more
techniques were introduced for using the feature-based cost estimation. Zhang et al. [42]
launched the feature-based CE by using back-propagation neural networks, especially for
packaging products. They introduced the definitions and the quantification of the cost-
related features, and then they used the back-propagation neural networks to figure out
the relationship between the cost-related features of product packaging and product cost.
Ou-Yang and Lin [43] suggested new tools for feature-based CE in the early stages of the
product lifecycle. They extracted factors that might affect production costs in an attempt
to reduce the costs in the design phase. At the end, their method can calculate the cost
based on products’ shapes and the accuracy of the extracted factors.
Leibl [44] designed a program-based system for feature-based CE. The program
integrates the CAD system to extract the geometrical and non-geometrical information
from the CAD model; next, based on the degree of details that were provided for the
program, the product’s cost is calculated. This system has to use other design document
to make a better cost calculation, as it cannot do a CE just based on the CAD model.
Brimson [45] used the feature-based CE as an alternative activity-based costing (ABC).
In this approach, the product’s features have to be identified first; afterward, the working
20
step or the requirement activity of each feature can be recognized. Finally, the cost of
each activity has to be calculated.
In the last two decades, many researchers have investigated into this area to improve
the efficiency and accuracy of feature-based CEs based on specific manufacturing
operations. F. Masmoudi et al. [46, 47] used a feature-based concept to do analytical and
parametric welding CEs with computer-aided tools. They described a CE model that uses
preparation features and welding features to approximate an effective welding time and,
thus, estimate the cost. They used the welding feature based on the forward reasoning
calculation with multi-parameter inputs, and they captured more details than they would
have taken with the traditional calculative approach. The limitation of this approach is
that to make a parametric CE, all of the feature data is required, which is not easily
available in practice.
In this thesis, a different approach is proposed using a hybrid method; in addition to
taking a similar parametric and forward-reasoning, CE route, a backward data mining
approach is also adopted as an alternative source of cost evaluation. Further, these two
routes are merged by a weighted combination for the ease of application and better
accuracy of prediction. The advantage of taking this hybrid approach is that historical
cost data and cost features are used and matched with data mining techniques, and the
backward approach could predict costs, even if all of the data has not been provided.
It is important to know that to enable accurate prediction of feature-based manufacture
cost, costs incurred should be related to the current plant capacity status and the forecast,
which can only be made available from the shop floor. Few CE systems consider this
issue [48]. In recent years, several CE models have been introduced to consider capacity
limitations [48, 49]. In addition, Wei and Ma [50] introduced the capacity feature concept
to fill in the gap between customer orders and resource management, which is considered
an important real-time input resource for cost estimation.
21
Besides selecting the appropriate CE approach, the cost estimation software is required
to improve the accuracy and consistency of the proposed approach. Currently, a large
portion of the software for cost estimation (SCE) are using data mining algorithms or
artificial intelligence (AI) techniques [51].
T.L. Karen et al. [52] reported about the effects of data mining techniques on cost
estimation; they described that data mining techniques are not always implemented better
than traditional cost estimation. However, a combination of data mining techniques can
provide accurate SCE for specific products or processes [52]. The goal of this research is
to introduce the combination of data mining algorithms that is suitable for feature-based
welding cost estimation, which can be used as a tool for ERP systems.
2.5 Review on software tools for cost estimation
Software cost estimation is an important domain in software engineering practice.
Different methods and techniques are using in software cost estimation. Jorgensen M et
al., [53] classified these techniques as a following order:
1. Regression: regression analysis is a statistical analysis that can shows the
relation among the dependent and non-dependent variables. Regression
analysis can be used in purpose of minimizing the sum of relative errors and
illustrating the non-negative coefficient [54]. Some cost estimation software’s
are using regression analysis for cost estimation. For example Constructive
Cost Model-COCOMO is one of the algorithmic estimation models that are
using basic regression formula [51].
2. Parametric: use the relationship between variables to calculate the cost
Analogy: Analogy cost estimation is a common method that uses similar and
completed product information to simulate the current product cost. The cost
related information of previous product is using as a basis for new product.
22
Also this method is using for validating the results of the other cost estimation
methods [55].
3. Expert judgment (EJ): based on the results of referring to the historical cost
data the software can do the cost estimation.
4. Work break-down structure (WBS): identify each process/activity that
generates cost
5. Function point (FP): the cost of single unit is calculated from past projects. It
is based on the number of functions the software has to fulfil.
6. Classification and regression trees (CART): CART is one of the most
important machine learning tools for constructing prediction models.
Classification tree is using specific rules for splitting data at a node based to
divide target variables into meaningful subsets and then regression tree is
using to predict the value in each terminal node [56].
23
Chapter 3: A feature-based semantic model for hybrid production cost
estimation*
3.1 Cost estimation model
Historically, attaining accurate cost estimation is a critical factor in business success,
because cost engineering has a great effect on the manufacture profitability and
management. Timely and accurate cost information helps business to make good decision
and reduces business risks. Also, accurate cost estimation can create a competitive
advantage. A good cost estimation method has a direct effect on sale price, amount of
sales, market share, and enterprise profits [12]. The cost management function can be
developed within the common Enterprise Resource Planning (ERP) systems. Those
successful companies that can offer the best price within the short time on market will
increase their market share. Lots of techniques and methods are introduced by researchers
for product cost estimation in recent decades. There have been number of researches
about cost estimation modelling and most of them focused on parametric and historical
data [57].
Feature-based cost estimation is an analytical cost estimation technique which is more
focused on quantitative approach rather than qualitative approach [12]. The main goal of
feature-based cost estimation is to calculate the product total cost based on identified
cost-related features. The most quoted limitation of feature-based cost modelling is the
difficulty to maintain the semantics cost related of features during the product lifecycle.
* A version of this chapter has been published in Journal of Engineering and Technology. Ma,
Yongsheng, N. Sajadfar, L. Campos Triana "A feature-based semantic model for automatic
product cost estimation".
24
To avoid this limitation, the complex cost-related features, and the relations among other
types of features have to be managed consistently.
In this chapter, cost feature has been defined [11] as a unique category of features in a
unified feature scheme [58] that can cover all types of features which are related to cost
estimation. As a result of associative feature modelling [59] which could support the cost
feature concept, then using just one kind of features generically so that it can be well
defined class to be semantically maintained during the product life cycle is a feasible and
effective approach. In such a way, the properties and behaviours of other extended
features defined on top of the generic feature model will be updated automatically. Figure
3.1 shows three sub-models, i.e. design model, machining model, and all other auxiliary
data including tangible data, which refers to the set of data that has been computerized
and can be extracted automatically by certain program interfaces, and intangible data
which requires user's input interactively on spot based on human experience and
knowledge. These three sub-models have conceptual semantic relations with each other's.
25
Cost features
Define the cost
feature scope
Define the cost
structure
Design
featuresMachines
features
Identify tangible
data and non-
tangible data
Mapping of cost
feature with design &
manufacture feature
Test the effect
of each feature
on cost
Determine the
effectiveness of
cost features
Combine all the
data with fuzzy
model
Test cost
estimation Data
Calculate cost
Suppliers’ data
Historical data
Marketing data
Licenses cost data
Product modules
Material
requirementWaterjet
equipment
Functional
assembly
feature
Surface
equipment
Part geometry
construction
feature
Functional part
feature
Welding
equipment
Welding speed
CNC equipment
CNC capacity
Cost feature library
Identify cost
features
ERP data
CAD/CAM data
product configurations
Machining
processes
Machining
features
Figure 3.1 Cost feature related elementary semantics with some non-exhaustive relations
26
3.2 Cost feature and cost feature association
The candidate contemplates to define a new type of features, that is Cost Feature, a
feature object class according to object-oriented software engineering paradigm, aiming
to cluster, encapsulate, update and manage those cost-related dependencies, rules,
constraints and referenced features defined from other domains, such as those related to
design functions and machining processes. The Cost Feature Association method is
introduced in this work to map the cost features with other types of features in cost
estimation via a set of associative and semantic trees.
According to the figure 3.2, different types of information is required For identifying
cost feature, such as the historical data, uncertainty factors, design and manufacture
conditions, etc. Knowledge-based techniques like Data Mining (DM) can recognize all
the product information from historical data, such as their attributes and the correlations
and relationship patterns among their attributes. The next phase is cost feature
establishment with attribute selection that is one of the most important steps. Given the
fact that not all the information of the product is useful for cost estimation, most effective
attributes have to be selected for this purpose [60]. Such selected attributes are then
associated by using a decision tree. In the next step, the rules for data analyses and weight
of importance are set. Finally, DM can present the data pattern that can predict the future
information from historical data. Some useful algorithms are defined, prototyped and
built into the methods of cost feature class as well as the derivatives of their object
instances.
Also, other supporting functions of the cost feature are developed to enable the
intelligent automation procedure and to persistently store information patterns as well as
the attributes that was discovered by data mining, such as manufactory specification and
inventory information.
27
Fuzzy model
Real world Virtual world
Data acquisition
System interface
Data synthesis
Cost estimation Software engineering
-Algorithms
-Organization
-Data organize
Data mining
Data base
Intangible data
Human input
Engineers
Designer
Extraction of
Tangible data
from all
documents
Analytical model
Semantic modeling
mapping
Real cost
dataCost
estimated
Cost
formula
Synthesis of
analytical and
experience models
Research,
Marketing,
Licenses,
Copyright,
Related
cost data
Comparison
Figure 3.2 Concept of semantic processing model for the proposed cost estimation
28
3.3 Semantic modeling
To explore the scenarios of applying cost features, semantic modeling is necessary.
Semantic modeling is a knowledge representation approach that expresses real world
objects in the form of predefined terms and entities that are interpretable by a computer
program such that the real world entities and their dynamic behavior can be consistently
described, modified and persistently reused within the prescribed scope of application.
Semantic modeling is an extension of high level object-oriented and modular software
development method that is useful for consistent software engineering modeling. It is one
kind of knowledge modeling that describes the meaning of data from the point view of
the computer entities with interactions and user interfaces.
The objectives of semantic modeling are abstracting the knowledge and understanding
the meaning of data and information. The other aim of semantic modeling is
implementing highly cohesive software engineering model [61].
Semantic modeling is useful for mapping the entities from real to virtual world;
modeling behaviors from complex life cycles of business, describing the dynamics of
activities and illustrating the processes so that well-structured software elements, e.g.
foundation classes, intermediate modules and functioning applications are modeled and
implemented by object-oriented programming. Figure 3.2 shows the overall concept
semantic modeling for cost estimation.
A disciplined semantic modeling approach for system implementation needs to follow
several rules:
1. All the steps of semantic modeling have to be defined clearly.
2. The abstract model should create relations for mapping the information between the
real and the virtual worlds.
29
3. The modeling objects have to be generically defined and directly mapped into class
diagram.
4. A semantic model has to specify the comprehensive relationships and interactions
among the modeling objects to describe the logics [11].
Three kinds of data are analyzed, classified, and related with some predefined
relations. The first kind is referred to as tangible data that are explicit attributes related to
cost objects describing properties such as physical, natural and economic data. On the
other hand, another kind, intangible data are those attributes that are not obviously or
directly related to cost elements but implicitly associated that the relationships are to be
discovered and defined dynamically like the data describing context, background, social
status, organizational attributes and cultural data, for example: copyright and licenses
have indirect cost for organization. The third kind of data is interactive and subjective
that requires engineering information and knowledge from marketing managers,
designers, process engineers, and purchasers; this kind of data fills the gap between the
tangible and intangible data types [62].
For doing reasonable and accurate cost estimation, all these three kinds of data are
needed. Tangible data can be materialized in manufacturing view with CAD/CAM data
models, historical data, ERP data, geometry data, manufacturing capacity, assembly and
completion data. Intangible data includes brand value, goodwill, training, skill
requirement, marketing positioning, software licenses, IP copyrights, research and
development cost, etc.
In the next step, the conceptual semantic relations illustrating the dependencies
between cost features and other engineering information entities are to be defined and
some elementary and non-exhaustive relations are shown in figure 3.1. Cost features,
supported with a template library, is a conceptual class of objects that has to define the
road map for doing the cost estimation within a specific scope. It has to define cost
constituents, searching mechanisms and target elements in the scope and a sound cost
30
structure. To interact with other types of features which are used for constraining the cost
elements and rules, a unified feature scheme can be used. Such a unified feature system
model covers a multi-face feature-oriented platform which defines the basic referencing
mechanisms and information inference scope which can include concept features,
machining features, user-defined face-cluster features, design features or any other kind
of features.
Also, it has to define the level of details that is necessary for features and the structures
that it wants to keep the feature dependency information. According to the scope and
level of features, then the cost feature can be defined by identifying appropriate attributes
and a set of constraints. Next, the tangible and intangible data sets are to be associated by
identifying their various referencing sources, for instance, from CAD/CAM data, product
configurations, ERP data tables as well as references to licenses, product models,
historical data and suppliers’ data. From CAD/CAM model and product configuration,
design features can be determined. A design feature includes all the information about the
material requirement, functional assembly features, functional part features and part
geometry constriction features.
On the other hand manufacturing costs can be determined from machining features
and machining processes. Machining features includes all the information about the
machines that are to be used to produce the specific product, such as a waterjet machine,
surface grinding machine, a welding machine, CNC turning or milling centers. Their
specific data sets can be easily searched and used such as welding speed, equipment
hourly rate and CNC programming capacity.
The most important step following is the mapping of cost feature attributes with design
and manufacture features and testing their cost effect by using a unified weighting
scheme. Then the variations of the effective cost features can be fully defined and the
total product cost and the price quote will be determined. At the end, all the data need to
evaluate the sensitivities towards changes of context and conditions as well as the
reasonability of cost estimation model for different cases.
31
3.4 Feature-based semantic cost modeling
A real industrial case of oil and gas equipment was selected to describe feature-based
semantic cost modeling. A Tong is commonly-used equipment for holds and provides the
torque to the drilling pipes on an oil rig. Figure 3.3 shows a Tong body assembly in a 3D
CAD model. According to the industry’s historical data, the safety door is a critical part
of tong body. Because the safety door is an important part of the product that the end
users will touch to open and close during drilling operation and to maintain the tong
body. The safety door is not a complex part but is welded together from six plates. Figure
3.4 shows the elementary plates of the safety door weldment, and table 3.1 shows the cost
of each part that was determine in manufacture level.
Figure 3.3 Example Tong Assembly Body 3D CAD model [86]
Safety Door
32
Legend:
① Latch (x2) ② Hinge (x1) ③ Spacer (x1) ④ Holder (x2)
Figure 3.4 Safety door model [86]
33
Table 3.1 ERP information about safety door part [86]
Item QTy Title Material
Cost ($)
Labor Cost
($)
Burden
Cost ($)
Time Spend
(min)
1 2 Latch 24.17 15.7 5.6 30
2 1 Hinge 8.7 12 1.14 20
3 1 Spacer 27 8.4 2.17 12
4 2 Holder 11 3.84 0.7 3
In the current costing system of the company investigated, the ERP information shows
the material, labor and burden costs for each item. However, there is no information
about feature cost from the ERP database. In this case, if the company wants to change
any of the design features, there is no systematic method to estimate the cost with new
feature.
For describing our new feature-based semantic cost modeling method, at first, the cost
feature has to be recognized. Figure 3.5 shows the parts and the face elements of the cost
feature for the safety door. For example SD1and SD2 in Part D can be produced by
cutting away the extra notches from the big blank part or welding the small blocks on a
smaller blank prepared; both options can be evaluated from the input data extracted from
the machining feature properties and using the relevant cost feature functions defined in
relationship to the critical face elements.
34
Legend:
SX#: Surface # of part X
Figure 3.5 Cost feature elements for the safety door module [86]
SA1SA2SA3SA4
SC2 SC1 SB2 SB1
mateoffset
Figure 3.6 Interfacing feature faces between parts A, B and C
35
Table 3.2 Cost feature information
Part Face Wi Thickness
(in)
Min.
feed
rate
(in/hr)
Quality Max. feed
rate
(in/hr)
Quality
A SA1 3 1.2~1.5 50 Good 75 Fair
A SA2 4 1.2~1.5 30 Good 50 Fair
A SA3 4 1.2~1.5 30 Good 50 Fair
A SA4 3 1.2~1.5 50 Good 75 Fair
B SB1 3 0.75~1 50 Good 75 Fair
B SB2 3 0.75~1 30 Good 50 Fair
B SB3 1 0.75~1 30 Good 50 Fair
B SB4 2 0.75~1 10 Good 25 Fair
C SC1 3 0.75~1 50 Good 75 Fair
C SC2 3 0.75~1 30 Good 50 Fair
C SC3 1 0.75~1 30 Good 50 Fair
C SC4 2 0.75~1 10 Good 25 Fair
D SD1 2 0.5~0.7 30 Good 50 Fair
D SD2 2 0.5~0.7 30 Good 50 Fair
D SD3 2 0.5~0.7 30 Good 50 Fair
D SD4 2 0.5~0.7 30 Good 50 Fair
Figure 3.6 shows that SA3 and SA2 have a critical relation with each other as a result
the weight of this relation is higher than the other features. SA2, SA3 is a hinge of control
36
for safety door. According to the interfacing feature relation schema, we can define the
weight as an attribute for the relevant cost feature [63].
For example the weight of SA3 (WSA3) is 4, because it has 4 connections with the
other feature. There for, any changes on SA3 has cost effect on 4 more features, however
any changes on SB3 just has an effect on 1 feature.
3.5 Conclusion
In this chapter, a new feature-based semantic model has been proposed for cost
estimation purpose. This model is built on top of three sub models: feature-based
association mapping, data mining and semantic modeling. Feature-based mapping model
is used to determine feature scope and cost level defined, including all the dependency
relations with other domain features such as conceptual, design, machining and user
defined face cluster features. Data mining was proposed to identify correlations among
tangible data, intangible data and human inputs for cost estimation purpose. Finally
semantic modeling used to map the cost feature attributes with other predefined features,
such as design and machining features. The chapter 4 will focus on mechanism
implementation and full details of the multi-face feature models with the proposed
automatic cost estimation model.
37
Chapter 4: Semantic Interactions within a Unified Feature Model for
Product Cost Estimation†
4.1 Design model
Refer to figure 4.1, in this research, the suggested design model is feature-based that
includes four sub-models: material requirement features, product assembly features,
functional part design features and part geometry manufacturing features. The design
model is created in the first phase of a product life cycle and enhanced or modified
throughout other downstream stages including manufacture stage. It can be appreciated
that the design sub-models product assembly features and functional part design features
have a significant role in cost estimation. The product design model is detailed gradually
based on the model's analyses by the way of further defining the model functions and
purposes of the product. Along the evolvement of design model, engineers should
consider processes of manufacturing the product. Generally, each design cycle has three
main stages: conceptual design, preliminary design and detail design [64]. Conceptual
design is about definition of product requirements functions, performance and possible
design solutions. During preliminary design a few solutions can be selected for more
analyses. Finally, the best solution will be selected at detail design stage.
There are several aspects for a well-defined design model, such as design for
manufacturing, assembly, lifecycle, recycling, cost and etc. [65]. Design model also
needs to present and incorporate the impact of manufacturing methods; hence design for
manufacturing (DFM) sub-module is necessary. Ideally it is a systematic approach that
considers the available facilities, tools and capacity in the stage of design product with
† A version of this chapter has been published in International Journal of Mechanical Engineering
and Mechatronics. Sajadfar, Narges, Luis Campos Triana, and Yongsheng Ma. "Interdisciplinary
Semantic Interactions within a Unified Feature Model for Product Cost Estimation."
38
features so that the product can be simply and economically manufactured. DFM is the
synthesis practice for designing component functions and strengths with additional
considering elements according to the manufacturing requirements. DFM module in a
computer integrated system describes operations such as welding, cutting and milling for
each component. For example, for a welding operation, it needs to describe the position,
height, length and weight of welding and all the drive path profile information. Note that,
the design model, based on the machining model, can enhance the product accuracy cost
effectiveness, and manufacturing flexibility. However, even in such a comprehensive
design model, it does not have enough input to include real-time machining information,
such as actual machine allocated cutting conditions, e.g. the feed rate. The designing for
cost is a common demand in the purpose of reducing product cost, but so far many
companies have difficulty to organize and extract the useful data for this advanced
application aspect.
The contents of design features change constantly during design process and so does
the development of sub-models. Therefore, associative design and manufacturing features
need to be used [58] to keep the consistency. The designer can use advanced feature-
based technique with a persistent design feature representation during the whole design
process. However, it is necessary to distinguish that design features are not
interchangeable with manufacturing features [63]. Figure 4.2 illustrates the model and
two views of a key component in an oil drilling equipment manufacturer, referred to
as door weldment as a real case study. Table 4.1 shows the member work piece
information of the door weldment. In concept design model, each member work piece in
the drawing can be defined as a weldment design feature. For example during conceptual
design door spacer is defined as a side impact protection feature. During the detail work
piece design, then the exact location, size and material of them are defined; and then
some material cost related data can be worked out.
39
Predefined cost
features library
Define the cost feature
scope
Define the cost feature
data structure
Design features
Identify tangible data
and non-tangible data
Test the effect of each
feature on cost
Determine the
effectiveness of cost
features
Combine all the data
with fuzzy model
Test cost estimation
Data
Calculate cost
Suppliers’ data
Historical data
Marketing data
Licenses cost data
Product modules
Material requirement
Waterjet
Functional assembly
feature
Part geometry
construction feature
Functional part feature
Cost features
Establish cost feature
object
ERP data
CAD/CAM data
product configurations
Product engineering data
Auxiliary data model
Design model
Machining model
Mapping of cost
feature with design
& manufacture
feature
Cost estimation
functions
Cost estimation module
Cost feature
internal methods
Cost feature tree
Machining
geometry
features
Machine
features
Machining
process
parametrs
Drive / guide
curve CNC turn
CNC milling
Welding
equipment
Waterjet
pressure
Travel
speed
Welding
speed
Machining
process
features
Target
feature face
Cutting paths
Identify cost feature
Cost estimation module
Cutter
Figure 4.1 Cost feature semantics with non-exhaustive relations
40
1) Plate door right hand 2) Plate door spacer 3) Plate door spacer 4) Post door
5) Plate door spacer 6) Door cylinder 7) Post door - Door weldment assembly
- Top view of weldment drawing - Side view of weldment drawing
Figure 4.2 Assembly drawing of door weldment and its individual work piece [86]
41
Table 4.1 Detail design information for work pieces before welding
Item Qty Work
piece
name
Purpose Weight
(LB)
Material
code
Top & Bottom plate 2 Plate Plate door right
hand
9.61 Medium carbon steel
Spacer 1 Plate Plate door spacer
among Top &
Bottom plate
4.27 Medium carbon steel
Spacer 1 Plate Plate door spacer 1.54 Medium carbon steel
Post door 1 Post Post door 1.34 Medium carbon steel
Spacer 1 Plate Plate door spacer 2.18 Medium carbon steel
cylinder 1 Lug Door cylinder to
make a hole
0.09 Medium carbon steel
Post door 1 Post Post door 2.01 Medium carbon steel
4.2 Manufacture model
The suggested manufacture sub-model in this research is also feature-based. Each
manufacturing process is embedded into feature which can be initiated from predefined
templates and fully generated with extracted feature entity information. For example,
machining feature associates the blank geometry, the related target part geometry, the
cutting geometry; the machines used, and detailed process conditions. This manufacture
sub-model has to represent the involved process data and entities via features; also, it has
42
to manage the relationships among the feature entities and the process data. Refer to [63]
for more details from both theoretic and application aspect.
Figure 4.1 provides the breakdown of the manufacture model, which has two main
parts: manufacture features and manufacture process. For example, a machining feature's
geometrical entities are generally defined as a set of faces and while it’s related tool path
driving curves are defined by referring to the part geometry as a set of derivatives; the
part and the derivatives are represented consistently with the Boundary Representations
(B-Rep) geometric model. Manufacture feature geometries are defined according to
associative features encapsulating the characteristic parameters of specific machining
processes, such as typical holes, slots, pockets, that can be identified from the model of
the final product [63].
There are several approaches for manufacture feature reorganization such as graph-
based, volumetric decomposition and hint-based [63]. Many researchers also have
attempted to define a standard set of machining feature, such as the Standard for the
Exchange of Product Model Data (STEP) that has been an international standard for
representing engineering data structure and exchanging product data model [58].
Theoretically, with the generic feature concept and further research work, user defined
machining features can be incorporated as well. However, cost information data structure
associated to design and manufacture features are not well defined so far. Table 4.2
shows the cost elements related to some manufacture processes for a door weldment. Five
different operations are needed and the machining features costs should be linked with
associated defined elements. Machining feature cost and its elements will be explain in
section 5.4 , the definition of welding feature as a case study of machining feature will be
describe with more details in section 6.1 and 6.2.
In addition, the machine configuration information such as equipment type, power,
dimensions, allowed interpolation, access and traversing directions, rotating, and
positioning attributes of each machine are modelled in the form of machine features.
Machine cost has to be considered as well.
43
Table 4.2 Manufacturing Information for the door weldment.
Operation
sequence
Operation
code
Operation
description
Work
piece
name
Run Run type
(hours or
minutes/
piece)
Setup cost
($/per hr)
Running
cost of the
machine
($/per hr)
1 Water jet Cutting Top &
Bottom
plate
1.3 Hrs/PC $30 $60
2 Man-Saw Cutting Spacer 6 Min/PC $30 $60
3 Drill & Tap Drilling Spacer 4 Min/PC $20 $50
4 Man-Lathe Turn/Face Spacer 9 Min/PC $30 $60
5 Weld Welding Door
Weldme
nt
3.2 Hrs/PC $10 $25
6 Mill Milling cylinder 0 Hrs/PC $30 $60
4.3 Auxiliary data model
Auxiliary data model is a collection of "design for x" data model under the concurrent
engineering concept, which supports product development and manufacturing aspects
with certain specific engineering and optimization. This model can be recognized as a
data association system that enables the sharing and updating mechanisms of system
integration across the product life cycle stages. For example, lots of tangible data need to
be associated for product cost estimation such as CAD/CAM data, product
configurations, profiling information, ERP data, available machines, supplier's process
and cost data, and product quality models. All of these data sources and uses need to be
identified at first step of auxiliary data model. Product feature library can define the
commonly used classes or objects information as templates to gather cumulative
enterprise knowledge and all the associated information accurately.
44
To introduce the proposed cost engineering feature model, a cost feature definition is
hereby given: Cost feature is a class type under object-oriented software engineering
methodology, representing an abstracted constituents of costing items in the form of
commonly recognized patterns, which contains the characteristic properties and
behaviour functions. The properties include characteristic attributes, geometrical entities
(including sub-features), referenced entities (including other features), etc. The
supporting methods of the class include creating, managing and deleting predefined
relations among the member entities, evaluation and validation methods of the attributes
and entities of the class; and the expected behaviour functions according to the generic
needs of cost engineering applications. Then individual cost features as objects can be
instantiated from the cost feature class to materialize individual applicable cost items.
Given the variety of cost items to be considered in engineering, cost features can have
different variations and they can be accommodated by applying object polymorphism
techniques. So are the cost feature member entities and attributes, specific sub-types can
be defined for different manufacture process categories
The main idea in this research is that cost feature member attributes and entities can be
identified and modelled with the help of data mining (DM) technique. To start the
description, the scope of cost feature data mining approach must be defined. It should be
started with the investigation of the scope of data mining applicable in this research field.
As we have discussed before, there are many sources of tangible data for cost
engineering. For example to set the price for an existing common part, we can usually
collect and refer to all the market prices. However for a brand new product, there is not
any market price and we have to look around for those similar products as well as the
benchmarking prices about the new product, and create a logical pricing formula the
reference to the elements of the auxiliary data model. However, product pricing is very
much dictated by the manufacturing costs as well, and they need to be worked out first
too. Hence, eventually cost features are useful for cost engineering throughout a product's
life cycle. To do that, different levels of information from different angles are needed;
45
hence, it is necessary to define product cost feature from deterministic relations among all
the related elements of any product in sales from raw material costs to delivery costs.
However, the proposed DM approach is based on another kind of data, which is
intangible data, which means the data has no clear (or too complicated) relations with the
targeted cost items. This kind of data has indirect effect on the cost estimation, such as
the computer software licences cost that used in design and manufacturing. It is
reasonable to assume that, after gathering all kinds of data with the help of modern
enterprise software tools, like ERP systems, CAD/CAM tools, or process planning tools,
the related cost feature member attributes and entities could be discovered by data mining
algorithms. Relating cost features with the relevant data set selected among the huge
amount of data is a challenging task that needs expertise.
4.4 Data obtained by employed sub-model
As shown in figure 4.1, the auxiliary data model contains ERP data, historical order
data with cost components, and product model configuration data. Obviously, useful data
can be extracted from such information repositories for cost estimation purpose. For
example design features can be analysed as they are commonly cycled in each product
order. Detailed design features can be searched and their pointers collected in a set of
attribute data structure because they have been fully implemented in the CAD/CAM data
model when the products were developed according to design feature templates. Further,
by searching the manufacture process model, the predefined process features can be listed
and tracked. However, to extract cost data from the ERP and order data models to
establish the cost feature objects, the cost feature templates, more research often has to be
applied. To do so, the cost feature library needs to be defined with the cost feature data
structure and the data extraction methods are required. It can be appreciated that typically
the three data models contain only past order data with those related design and
manufacturing features while the explicit cost relations are not yet identified.
46
To solve cost feature identification problem, the historical order data with cost
components from ERP data model has to be analysed and associated dynamically to the
design and manufacturing features in order to justify the estimations and also directly
support the functional estimation module. Therefore, the next step is to develop the
feature mapping relations based on the results of the above data extraction procedure and
identify those input relations that will be used as cost feature mapping technology.
Up to now, several methods were introduced for feature mapping, such as
mathematical model [66]. To keep the implementation simple, the mapping matrix
method suggested by Zhang et al. [67] is used for implementation in this work. The
outcome of mapping matrix can illustrate the cost estimation function. Mapping matrix is
sufficient to manage and transfer different cost features, however it cannot cover all the
relation types or details. By using the output function generated from the mapping matrix,
the cost features are identified for each product cycle; and hence the accuracy of cost
estimation will be increased.
4.5 Data mining for cost feature pattern construction
Clustering and classifying the data can be useful for data management. The first step is
clustering the available data according to application scopes. This step can be done in two
ways: clustering and classifying according to the assembly principles of component
analysis, or using a predefined clustering algorithm. Then, within each cluster, DM
process can be used for further cost engineering. By applying DM in a cluster with the
appropriate size, the accuracy of data mining results will be increased, because there is a
set of common relations among the components of each product family which has similar
data patterns. Within a cluster, components will be classified by the same decision tree
and rules.
47
The second step is to classify cost related data among components in a cluster. During
the DM classification process, two kinds of data will be encountered: quantitative data
and qualitative data. Quantitative data can be classified by a decision tree via rule setting.
Linear regression, K-nearest neighbour and baseline predictor are some common
algorithms that can be used for analysing qualitative data.
In the next step, by analysing the combination of classified quantitative and
qualitative data, the cost model patterns will be extracted. Model patterns can be used to
filter the selected cost related data and as a result the cost feature constituents can be
established with validated DM processes. At the end, these constituents, in the form of
attributes or pattern data structures are clearly identified and created into reusable class
elements. These elements are then further ranked according to the nature of the product
that the cost estimation method will be applied. Also, there are several data mining
algorithms reported for data ranking. Such as, best first (find the best data first), linear
forward selection and ranker (find the best set of data), which can be useful for data
ranking. Then, one of the most effective ranking techniques can be selected based on the
weight of data or accuracy of data [68].
Figure 4.3 shows the steps of auxiliary data modelling with more details. Table 4.3
shows two resulted patterns generated based on cost related data that can be used for the
estimation of cost for producing door weldment. Note that the simple average cost per
part accounting all the cost elements should be $831 for alternative 1 while $518 for
alternative 2. However, the resulted average cost per part after data mining is not exactly
the same. They are $787 and $483 respectively. The differences reflect the influence of
intangible data on the final cost due to the inconsistent data sources and discretion by the
management. The interpretation and accuracy evaluation for the data mining method are
to be further studied in the future. The system design details are to be introduced in the
next section.
48
Tangible data
Intangible data
Identify auxiliary
data
Selecting cost
related data
Quantitative data
& geometric data
Qualitative data
Classifier
- Decision tree
- Rule setting
Model patterns
DM process
Cleaning the data
Valid cost related
data
Effective ranking
Select the best
valid cost related
data
Figure 4.3 Auxiliary cost modeling with data mining process.
Table 4.3 Generated output by integrated model.
DM method Machine
cost/yr ($)
License
cost/yr
($)
Maintenance
cost/yr
($)
Labor
cost/yr
($)
Material
cost per
part
($)
Output
part/yr
Average
cost/part
($)
Quality
Alternative
1
420,620 25,000 420 1,600,000 45 2600 787 Good
Alternative
2
320,000 25,000 360 1,200,460 35 3200 483 Fair
49
4.6 System design for cost estimation
As explained, having a unified cost estimation system is imperative for doing dynamic
cost engineering throughout the stages of a product lifecycle. Figure 4.4 illustrates the
computer system design for cost estimation conceptually with reference to the modules of
an ERP system in use in the sponsoring company and those related product engineering
models. Cost estimation system begins with the customer ordering process. The customer
requirements are first collected, processed and recorded into the ERP customer relation
management (CRM) module. Then such requirements are mapped into
product functions which are defined by the product model series, and up into a customer
specific product configuration [50]. Based on such a product concept model, unified
feature model is established which includes design features, manufacturing features and
the corresponding cost features. Then, with the relevant information provided from the
product model, a cost advisory system module needs to manage the cycles of "if-then"
design analyses about the costs. Via the cost advisory module UI that shows in figure 4.4,
the interactions between customer requirement and cost engineering, modules which are
supported by the organized cost features, can be realized dynamically.
From the synthesized cost results which are functionally supported by the cost
advisory system, the customer is then informed on the estimated product cost much more
accurately than using traditional experience-based rough estimation. After, cycles of
evaluation on the product in contrast to the cost specifications of the product expected,
the customer can make then the purchasing decision; and a customer order is generated.
Once the customer order is confirmed, from here the cost management process kicks
in. The cost management system is activated and the corresponding UI as described later
in this chapter shows the actual cost components of that customer order. In the next step,
all the cost information from cost advisory system will be transferred to cost management
module and it will also be used in the ERP system. Cost features are used in mapping cost
50
related associations by referring design features and manufacturing features under the
product unified feature model.
4.7 Integration among sub-models
Each of the above data models can be defined as a module in the system and mapped
by a functional modular class; i.e. design class, machining class and auxiliary data class.
Semantic data modelling between these three classes provides the integrated system
within different views on data; such as structured and non-structured data [69]. To realize
the expected functionality, several types of class relations among sub-models can be
defined. A relation can be defined as a semantic link between two or more objects for the
purpose of creating a logical system. Conceptually, a relation can be defined as an
attributes, entity, independent element or function [70]. In the sense of object-oriented
approach taken by this research work, the relations among classes can be defied as an
aggregation, inheritance, using, association, and instantiation [71].
Figure 4.5 shows the UML diagram of integration with sub-models. Cost features are
associated with all the three sub-models, or the three modular classes. First, there is an
association relation between design feature class and machining feature class. Due to the
interlocking nature of design and manufacturing considerations in real product lifecycles,
there is some kind of mutual dependency link, cooperate, between these two classes.
Auxiliary data model is an abstract modular class that has “reference” relationship from
the design feature class and the manufacture feature class. It means auxiliary data class is
dependent on design feature class and machining feature class; and any change on the
design and manufacturing will be automatically reflected on auxiliary data class
properties. Cost features refers data input from all the three sub-models; but the real
relationship materialization can be constructed based on the input data and the data
mining process results as discussed in section 4.6. Therefore it is clear that cost features
51
are modelled to take care of the intricate cost influencing relations initiated from design
and machining features as well as other auxiliary data.
It can be appreciated that usually, the features that used for design have different
definitions from those related manufacture features. To clarify the interdependency
relationship between design and manufacture features, more discussion is necessary.
Figure 4.6 shows a part of door weldment and their geometrical elements of design
features and the machine features on this part respectively. In design geometry, the part
was defined by adding a small block on top of base block and cutting a curved notch
from bottom of base block. On the other hand, the part machining process was defined by
water jet cutting represented with machining feature profiles B, C and D.
52
ERP
DB
Customer
order
Customer
requirements
Cost advisory
system
Cost
management
UI
Advisory
system UI
Function
definition
model
Product
configuration
model
Manufacture
features
Design
features
Mapping of
cost feature
with design &
manufacture
feature
Product unified feature
model
Cost
management
system
DeliveryFinish good
storage
Outsource
partsWork orders
In-house
assembely
plan
Outsourcing
cost centre
In-house
machining
Part supliers Schedule Materials
WorkersWork in progress
storage
Cost features
ERP system (partial)
Product
DB
FeedbackProduction
planning &
scheduling
Product concept model
Customer
Cost engineer
Information flow
Figure 4.4 Computer system diagram for cost estimation
53
-ERP data
-suplier data
-historical data
-licenses
Auxiliary data
-title
-qty
-material
-subject
-function
-configuration
Design feature
-operation
-equipment
-capacity
-setting info
-feed rate
Machining feature
Cost related attributes
Cost feature
Inheritance
relationship
Cooperate association
relationship
-Cost methodsManufacturing
feature
Figure 4.5 UML diagram of integration between sub-model features.
Geometry design features Base block A + Protrusion B – Curved notch C
Machining features Base block A - water jet cutting profiling features B, C, D
Figure 4.6 Difference between machining features and geometry design features
54
4.8 Conclusion
In this chapter, the author defines a cost feature class, and then presented an overview
of the relations between cost features and three engineering sub-models, i.e. machining
model, design model and other auxiliary data model. For having the dynamic, unified and
accurate cost estimation modelling, the relations among cost features and all engineering
sub-models need to be managed consistently. The descriptive semantics of cost feature is
flexible enough to represent all types of associations that are related to cost estimation,
having deterministic and non-deterministic. For the latter type of relations, data mining is
proposed as a useful method to extracting data patterns for these three sub-models. It was
also discussed that clustering before data mining is helpful to assure the quality of the
approach. Classifying cost feature elements according to data mining algorithms was
proposed for intangible data analysis in the purpose of more accurate cost estimation.
Currently, the data application and error evaluation have not been carried out yet. They
are two important aspects to proof the proposed algorithm; and they will be addressed in
the following chapters.
55
Chapter 5: A Hybrid Cost Estimation Method Based on Feature-oriented
Data Mining ‡
5.1 Current practice of cost calculation in ERP systems
Due to the increasing manufacturing product complexity, variations, as well as
dynamic supply chain management, manufacturing companies need to coordinate its
order acceptance process dynamically hence they must make decision based on timely
information resources among their business partners. The customer inquiry management
has to be done accurately and quickly to provide effective feedback.
Currently, different computer-aided systems are used for collaborative manufacturing
that aims to increase efficiency and flexibility while also keeping sensitive collaboration
loops with the different business partners. Enterprise resource planning (ERP) is one of
the most important management order-fulfillment tools capable of creating the links
between product documents, schedules, and other forms of communication. ERP
software is used to share information among different partners for daily manufacturing
activities, through which many of the process functions are intended to support
collaborative manufacturing, such as customer ordering, process plans, financial decision
making, accounting, and supply chain coordination, all of which are unified and managed
from a single system [72].
However, ERP packages currently do not have the sufficient functions necessary for
market-oriented cost estimation with dynamic marketing and product configuration, even
though such functions would play a major role in business success [73]. Cost estimation
‡ The major part of his chapter has been submitted to Advanced Engineering Informatics, by
Sajadfar, Narges, and Yongsheng Ma.
56
(CE) is one of the key success factors that create integration between in-house business
and market-oriented functions.
Manufacturing costs in many companies are not accounted for systematically and
accurately due to the complexity and the constant changes of the processes involved in
production and the lack of data collection schemes. The process model in an ERP
package contains certain details of manufacturing processes, such as the types and
setups of machining features, cutting conditions, expected productivity, past costs, and
status of job completion. As a result, in this work the process information is integrated
within ERP for CE purposes. However, one limitation of most ERP systems is that the
actual variations of processes, such as the different welding jobs, are not well defined and
documented. Such technical challenges are particularly true in companies with small
batches and high variations of product models, which cause overwhelming difficulties in
accurate CE. Developing a generic and comprehensive data model for manufacturing
processes in ERP has been a major issue that is being addressed to achieve the expected
production performance [73]. Most ERP systems can estimate process costs based on the
required time for the machining process (i.e. labor cost), materials, and power
consumption. However, usually, the complete information for a specific manufacturing
process is difficult, and making use of the historical data from similar operations requires
an expert judgment to determine the factors that correlate to the respective costs.
Traditionally, the cost of a product is calculated based on the time spent on product
development, the manufacturing processes required, equipment used, and the energy and
materials consumed. In turn, these are influenced by the parts’ shapes and characteristics.
On the other hand, the modern cost paradigm focuses on customer inputs, product
similarities among configurations, and flexible markups [74]. Beside cost calculations,
there is a considerable amount of research available on CE for manufactured goods,
specifically about the manufacturing processes, such as turning and milling CE.
However, many process costs have not been clearly defined yet, such as the cost of
welding operations in variation workshops. Analytical and parametric methods are
57
commonly used for manufacturing CE. Such methods are difficult to adapt to different
jobs, and estimation can only be given by experienced personnel [73].
5.2 ERP solutions
ERP vendors are developing stronger financial functions for cost management. A
typical ERP system records and tracks product and process related data such as raw
material, set-up time, run time, number of machines and labor time for cost calculation.
The disadvantage of this method is that many detailed features and parameters are not
considered in cost calculations; therefore, the accuracy of the cost results is questionable.
In addition, many cost items can only be calculated when all of the parameters are in
place, this constraint requires more time and data or even makes timely cost estimation
too expensive to be implemented
To solve such problems, ERP vendors are developing accurate cost forecasting
methods. For example EPICORTM
[75] is working on the development of cost forecasting
and planning in addition to making calculations of manufacturing costs. In the EPICOR
TM system, the budget process is defined based on the manufacturing process plan in
multiple cost-calculation scenarios. Then, the financial planner uses different methods,
such as examining history and trends and making a depreciation analysis, to do the
forecasting [75].
Currently, in an ERP system like Epicor, the parametric CE is still the basic method
for quotation generation and final cost determination for a company’s manufacturing
products and processes. Usually, the ERP system conducts cost calculation based on four
categories: labor, material, burden, and service, which can provide the general
information necessary to determine cost. To have a more accurate CE and calculation,
more attributes needs to be extracted. As an example, figures 5.1 and 5.2 show the cost
58
attributes of a specific part (Part ID: 1095-139 from studied company) in EPICOR and
[76].
Figure 5.1 Typically-defined cost attributes in EPICORTM
Standard cost
that defined by
company for each
category
Last cost for each
category based
on historical data
Average cost for
each category
based on
historical data
Fist in first out
average cost for
each category
based on
historical data
The unit of cost is $
First-In, First-Out (FIFO) is one of
the methods commonly used to
calculate the value of inventory
The part that selected
59
Figure 5.2 An example set of cost attributes defined in Visual InfoTM
5.3 The principle of hybrid cost estimation method
A hybrid CE method includes two main modules: empirical analysis and DM analysis.
Historical data and cost feature libraries have been used as input for these analyses. The
initial input of the CE system is the historical data, which is usually available in many
companies but in different table forms. To unify the data structures and create a
characteristic representation of cost components and calculation algorithms associated, a
The part that selected
The unit of cost is $
Average cost for
each category
based on
historical data
Total cost Method of
manufacturing
(MOM)
Bill of
materials
(BOM)
60
new concept of the feature-based CE model has been introduced by authors [58, 77]. The
authors have defined Cost feature as a unique class in the unified feature modeling
system to address the characteristics of cost-related data, restrictions, and dependency
relations, which are closely related to the product configurations and the manufacturing
processes involved. By using the product and manufacturing process information with the
support of historical data information, cost features are extracted into consistent data
structures, and templates are created to assist complete data gathering. They are
organized into a pre-defined cost feature library, making it available for the feature-based
CE. The cost feature class diagram is introduced in section 5.4.
With the sub-classes of the cost features, all components can be divided into different
categories according to their principal functions, geometry information, similarities, and
process plans. As a result, the feature library can classify common-feature families into
the same class. It can also use a feature-mapping technology [66, 78] to create a
classification for each manufactured product. The proposed method uses pre-defined
feature objects to create links between each feature template (class) and its instances
(details of costs for the data items).
In this proposed system, to support the hybrid approach, a CE engine has been
constructed with the inputs from the historical data and the cost feature library, as figure
5.3 illustrates. An empirical cost analysis method has been developed based on the
approach suggested by Masmoudi et al. [46, 47]. Regression analysis with expert
judgment is carried out to develop a feature-based cost calculation formula. As figure 5.3
shows, an empirical analysis is one of the two complimentary CE methods to estimate
product costs and their accuracy.
In contrast, in another branch of the hybrid system, the DM software algorithms are
applied to the sample historical data to analyze and extract the pattern for each data set.
After evaluating the cost patterns, the DM module will estimate the cost components and
their accuracies for new product inquiries.
61
Empirical analysis DM analysis
Regression
analysis
Trend
analysis
Expert
judgment
Cost feature
formula
Select
sample
Data
clustering
Apply DM
algorithms
Evaluate and
visualize
pattern
cost Accuracy
Weighted cost
estimation module
Historical data
Weight
determination
Historical
data
Cost feature
Library
17$ ±0.03
cost Accuracy
17$ ±0.05
Expert
judgment
Cost
resultAccuracy
Report outputProgressive
real data
Compare
Information repository
Validate &
update
Auto accuracy
evaluation
Feedback
Cost estimation engine
Interactive synthesis module
Figure 5.3 A hybrid cost estimation method proposed
62
These two CE methods are validated with the historical data, and they are combined
with a weighted sum mechanism, as figure 5.3 shows. The weight given to each method
is based on expert judgment from the validation results or is suggested by a system’s
accuracy evaluation algorithm. The authors believe that the proposed hybrid method can
achieve accurate CE with adaptive adjustment capability in real implementation.
5.4 Definition of the cost feature
As described in previous chapters, a cost feature is a set of an object’s characteristic
attributes and their related functions and servicing methods, which can be associated and
constrained to represent the pattern of semantics for the purpose of cost engineering. The
cost feature has to consider the variety of cost-related information: quality; operation
sequence and time; design; equipment; materials; functions of application; and geometry
and non-geometry details, such as tolerance, surface finish, and working allowance. Also,
each of these factors has its properties and behaviors that must be taken into the
consideration for cost feature implementation and applications. Figure 5.4 illustrates the
partial relations among different features for cost engineering, in which the process cost
feature and product batch cost features are presented.
63
Figure 5.4 Cost feature definition in UML format
64
For extracting the cost feature, we need to provide reference data that includes specific
design and manufacturing information, such as the number of available machines in the
shop floor, the capacity of each machine, the operation that can be performed by each
machine, the available human resources that can work with each machine, and the
sequence of the machining process. Then, some functions are required to extract the cost
features and related information from the design feature model, manufacturing process
features, and ERP system.
According to [43,46] for a welding process cost feature, which can be a child class of
the process cost feature, as shown in figure 5.4, the following welding feature class is
suggested:
Where constraints means:
- (): welding length
- SS (): welding section
- (): the rod output
- (): the filling metal density
- (): the process efficiency
- (): protection gas-consuming time
- (): operator efficiency
- (): degree of weld position complexity
- (): unit time cost ($/hour) for machine h
- (): setup cost ($/hour) for machine h
Where functions means:
- (): welding volume evaluation: = SS x
65
- (): amount of welding wire:
- (): electrode time:
- (): welding operating time:
- (): machining cost for welding:
Table 5.1 Cost feature template
Feature
ID
Feature
type
Process
step
Process
type
Process
description
Welding
sections
Welding
operating
time
Machining
cost
Material
cost
xx-01 T-joint 1 Shielded
metal arc
welding
End to end
with vertical
borders
(e,
Regarding extracting information from parts’ geometry and process features, some of
the weld information has to be inquired from the CAD model so that the welding time,
welding length, volume of welding wire (rod), and the welding tool as well as its control
can be worked out.
As table 5.1 illustrates the cost component variables identified in the cost features are
to be used as the initial input for both the empirical estimation as well as the data mining
(DM) analysis. Although it is ideal to achieve the above-mentioned data extraction by
automatic algorithms embedded in CAD/CAM tools, engineers’ interactive input through
friendly and consistent user interfaces can also be accepted. Hence, the application barrier
for this proposed cost-estimation method can be significantly lowered.
66
5.5 Empirical cost estimation process
Usually, companies use the parametric cost calculation formula to determine the cost
of each product. However, they are not considering all of the products’ features when
they use this formula. As a result, their cost calculations are not accurate. To solve this
problem, we used empirical analysis to become confident that we considered all of the
products’ features to achieve an accurate CE.
Figure 5.6 shows the proposed empirical CE process built into the CE engine. The
empirical CE module uses the existing ERP data and interpolates data to analyze the
weights and coefficients of cost component initial variables from the cost feature input.
Regression analysis is a useful tool for evaluating the weights of variables and the
coefficients among them. Regression curves can show the data strength and their
relations. By using the above analysis, the relative statistic cost weight predictions and
coefficients can be extracted out of the initial variable set; afterward, those contributing
and independent variables are quantitatively used to form a linear regression CE formula.
A cost data set is defined as { } where the total cost of the product is
denoted by ; it is a differentiable function of input component variables:
.
The goal of regression is to draw a line through the data set that can pass data as much
as possible. To find the best-fit line or curve of the data, the regression has to analyze the
relationship between the variables that are denoted by to illustrate the coefficient
among and , which means how changes in can affect .
(5-1)
67
We need to find the best set of to minimize the error and build a more accurate
formula for CE. To achieve this goal, the data related to each variable has to be analyzed
to qualify the product data with high precision. For example, observing the amount of
welding wire consumed in 40 products and their relation to total cost is shown in figure
5.5. After analyzing the data of each variable, we can use its coefficient to build the CE
formula.
Figure 5.5 Relation between total cost and amount of welding wire
The CE formula needs to be updated after the evaluation with the existing pool of the
existing data, and then, it can predict the initial cost and its accuracy for a new product or
order. Along the way, the estimation confidence level is constantly checked. If the
predicted costs cannot meet the confidence level required, the weights of the regression
analysis have to be updated with regression refinement. This process will be iterated
several times to ensure that the CE formula and the level of accuracy are acceptable.
0
2
4
6
8
10
12
14
16
18
0 0.05 0.1 0.15 0.2 0.25 0.3
tota
l co
st $
Amount of welding wire (kg)
Y
Predicted Y
Linear (Predicted Y)
68
Data sampling for
pre-estimation
analysis
Qualification
weight analysis &
Regression
InputERP
data
Interpolation
data
Regression analysis
Building
estimation formula
Evaluation of
formula
Feedback
Predict
initialize cost:
17$
Predict accuracy:
±0.05
Establish final cost estimation
Update weight
Output
cost Accuracy
17$ ±0.05Meeting
confidence level
requirement
Yes
No
Example:
...
Data repository Empirical cost estimation module
Figure 5.6 Empirical cost estimation process
Five hundred products were selected as sample data elements for testing regression
analysis for the purpose of CE. The output of regression analysis between the total cost
Ctotal and { Xi } shows that Multiple R (correlation coefficient) is 0.99, which is high. The
standard range of Multiple R is between -1 and 1, in which 1 is showing a strong relation
among the variables. In addition, R Square (coefficient of determination) is 0.97, which is
also high. The range of R Square has to be between 0 and 1, and 1 means that the
regression line passes exactly through the data. Figure 5.7 shows the scatter plot of the
multiple regressions that have been generated by regression analysis. Note that the
products have been organized according to the individual total costs.
Table 5.2 Regression Statistics Result
Multiple R 0.99
R Square 0.97
Adjusted R Square 0.85
Standard Error 0.33
Observations 500
69
The formula extract from regression:
Welding cost = 1.2812E-13 + (Labor cost per hour -0.059562) + (Amount of welding
wire 1.44) + (paint cost 1) + Gas volume rate + (Machine cost ) +
(overhead cost -0.621) (5-2)
Figure 5.7 Sample of scatter plot of the multiple regression analysis
5.6 DM cost estimation process
An accurate CE can be seen as an optimization problem in which we have to search
for the most accurate estimate. Data mining is a powerful technique that can be used in
CE. Indeed, there are many data mining algorithms, and each is suitable for one kind of
data modeling. Some widely used techniques such as linear regression can describe data
coefficients and behavior.
Data mining can extract knowledge from a significant amount of data, transfer the new
data into useful information, and discover useful information and hidden patterns by
0
2
4
6
8
10
12
14
16
18
0 10 20 30 40 50 60 70 80 90 100
tota
l co
st $
product
70
analyzing the data. In the recent decade, data mining has been widely used for data
training and testing in order to extract data patterns.
The proposed DM CE method has several interfaces with the product domain and the
process domain. Figure 5.8 illustrates the DM cost estimation configuration. The product
domain contains information about design configurations, modules, parts, design
features, and materials. The process domain defines manufacturing processes, such as
machining operations and their related machines, tools, and labors involved.
On the right side of figure 5.8, the DM cost estimation module is shown. It can
calculate the cost of each specific manufacturing feature based on the information of the
product and process domains as well as the ERP historical data. To achieve this goal, the
proposed DM module uses ERP historical data to categorize cost data by product families
based on product configurations, materials, and manufacturing processes. In the next
step, the DM module analyzes the cost data to associate the cost values with cost feature
variables. Then an appropriate data mining algorithm is selected and used to extract the
full cost feature patterns (i.e. object instances). As a result, the selected data mining
algorithm is associated with the sample data product family. In implementation, the user
interface will show the confidence level of each algorithm to the user, and based on that,
the user can select the best algorithm for each product family (i.e. group).
71
Figure 5.8 DM cost estimation configuration
ERP
Historical
Data
Product model Modules
Assembly
Part
Manufacture process
Data Mining
Cost feature
result
Estimate cost
Evaluation
Cost resultConfidence
level
Extract
Manufacture
features
Raw material
Machine
Tooling/ Tools
People
Cost pattern
parameters
Accuracy estimation
BOM
MOM
Design
features
Data Mining
software
Cost feature
library
Discover
feature
Methods &
algorithms
Conceptual modules
Member
Product domain Process domain
DM cost estimation module
Functional modules
Member
Associated
Info share
Map to
Legends: UML
72
5.7 Data mining implementation in the cost engine
For implementing the proposed DM cost estimation method, the cost-related data has
been saved in the database. Several data mining algorithms have been implemented and
tested in WEKA to extract and process this significant amount of data in an efficient
manner. WEKA has been selected in this research because it is machine learning software
that provides a comprehensive collection of machine learning algorithms for automatic
classification, regression, clustering, and feature selection [79].
Usually, SCEs use different algorithms and Artificial Intelligent (AI) techniques. Such
as Linear regressions, multiple regressions [80], and artificial neural networks (ANNs)
[81, 82] which are commonly applied in many types of CE software. They reported that
these algorithms are more suitable for numerical data, and ANN presents more accurate
data with a lower error percentage [80]. Besides the previously mentioned techniques, we
added KNN and MP5 data mining algorithms, which are suitable for both numerical and
non-numerical data and multiple relations [83].
In the next step, the results of the DM algorithms were used to generate new rules and
patterns for welding CE. Then, the result of the algorithm was saved as a model to apply
to cost estimation. Further, the user has to select the most effective CE algorithm based
on confidence level. Confidence level can be defined by comparing actual effort and
estimated effort based on the mean magnitude of relative error (MMRE). MMRE is a
standard function used in SCE that is applied to evaluate the performance of CE on a
training data set [53]. Where n denotes the number of manufacturing parts used for CE
and MRE is equal to the following:
MMRE =
(5-3)
MRE =
(5-4)
73
Chapter 6: Case study – welding process cost estimation
6.1 Definition of welding process features
Welding features are a new type of manufacturing features defined in this work, which
have not been formally defined in the literature thus far. Figure 6.1 shows a typical
welding feature that currently uses. In this work, the welding feature is a child class of the
manufacture feature (see Figure 6.2) , and it contains a set of extracted welding related
process data, which is associated with welding geometric entities and their characteristic
parameters, technical process setup parameters, and related resources. Welding-related
geometric entities are further related to product design features and manufacturing
derivatives, such as welding paths and which angles have to be defined within any
welding feature.
Figure 6.1 A typical welding feature
Work piece A
Work piece B
Welding seam
Driving path curve
Welding rod
Welding torch Welding jaw
74
Referring to the welding feature class as shown in figure 6.2, the welding feature
operation process is executed via three main predefined class methods: preparing-
welding(), run-welding(), and after-welding(). They interact internally with the feature
properties as shown in the figure. For example, the preparing process is defined with the
setup information; it extracts technical parameters from the design features, related
manufacturing features, and generates the necessary drive curves and path segments as
well as characteristic geometric parameters from product design and structure entities.
Note there is a unified product feature model shown in figure 6.2. The welding process
planning function uses the manufacturing features from the product unified feature model
as well. The run-welding() methods uses the real-time information of the assigned
welding equipment resources from workshop process planning and operation model via
their data object pointers, such as those describing welding machine and welding tools.
Finally, after-welding() methods is supposed to define and guide operational application
to do cleaning, inspection, and test tasks.
75
Manufacture
features
Design
features
Mapping of
cost feature
with welding
feature
Product unified
feature model
Cost features
Properties:
- Technical setup parameters
- Geometric parameters
- Design & structure
- Welding drive curve/
segments and path
- Welding control parameters
- Quality requirement
Welding features
Welding process
planning
Run-welding process
After welding process
Welding
operation
Resources:
- Welding machine pointer
- Tooling fixture pointer
- Rod pointer
- Robots pointer
Preprocess groove
preparation
Cost estimation
engine
Application methods:
- Initialize ();
- Develop_setup ();
- Develop_weld path ();
- Generate_control paths ();
- Select_weld rod ();
- Simulate ();
- Calculate_time ();
- Calculate_cost ();
- Prepare_welding ();
- Run_welding ();
- After_welding ();
External processes:
- Call cleaning ();
- Call inspection ();
- Call test ();
Use
Info share
Legends:
Manufacture feature classWelding feature class
- Type of operation
- Geometric parameters
- Equipment requirement
- Quality requirement
- Safety requirement
Related
operations
Workshop process
planning & operation
model
User interface
(UI)
ERP data
Instantiate
Cost estimation
module
Figure 6.2 Welding feature and cost feature structure relations
This case study has implemented a preliminary CE engine that works with welding
features whose template is defined as a feature class. In addition to the live support from
the instances of functional modules as indicated in figure 5.8, the feature library will
76
collect all the information related to welding features to classify, sort, and analyze them
for future data mining processes.
6.2 The data structure of welding feature
In this research, the following data structure of welding feature class is suggested:
- L1, W1, H1: length, width, and height of first piece
- L2, W2, H2: length, width, and height of second piece
- T1: time of weld
- T2: time of setup
- T3: time of paint
- T4: time of clean
- BM: Base Material
- WThick: weld thickness
- WType: weld type
- Tech: technology used in welding
- WPosition: welding position
- WProcess: welding process
- WW: amount of welding wire
- PP: amount of protection paint
- GT: gas type
- GV: gas volume rate
- OF: operating factors
- StartP: the location of start point
- EndP: the location of end point
Figure 6.3 Data structure of Welding feature class
Data structure of welding feature class
Properties:
- Geometry information of first work
piece: L1, W1, H1
- Geometry information of second
work piece: L2, W2, H2
- Welding times: T1,T2,T3, T4
- Technical parameters - Welding geometric parameters
……………
Constraints:
- L1 > 5cm
- 90° < α <180°
- WThick < 10 cm
…………….
Functions:
- Get-parameters ();
- Save-parameters ();
- Calculate-welding-time ();
- Initiate-feature ();
- Initiate-UI ();
……………….
77
- form: form has three positions: End to End in V; End to End with vertical borders; and
End to End in X. Figure 6.4 illustrates these three positions:
Weld feature objects Diagram
End to End in V
- e: height of thickness
- t: heel or height of the flat
- α: chamfer opening top angle
t
α
e
End to End with vertical borders
- e: height of thickness
- g: clearance apace g
e
End to End in X
- e: height of thickness
- α: chamfer opening top angle
- β: chamfer opening bottom
angle
- g: clearance apace
- t: heel or height of the flat
gt
α
e
β
Figure 6.4 Geometric Parameters of welding in different welding form [46]
According to the current definition, P1 represents the geometric information of the
first piece such as length, width, and height. P2 represents the geometric information of
the second piece that needs to be welded to the first piece. TP includes the technical
parameters of welding, which includes base material, weld thickness, weld type,
technology used in welding, welding position, welding process, amount of welding wire,
amount of protection paint, gas type, gas volume rate, operating factors, the location of
start point and end points of welding. GP is the geometric parameters of welding, which
includes shape characteristics. Table 6.1 illustrates the data structure of the welding
feature properties:
78
Table 6.1 Data structure of welding feature properties
welding feature
properties
abbreviation Unit
Base Material BM Inch, SQIN
Weld Thickness WThick mm
Weld Type WType Grove, Seam, Joint
Technique of Grove Tech Grove U, square, flare, bevel,
V
Technique of Seam Tech Seam single side, double
side, all round
Technique of Joint Tech joint Square butt joint, V
butt joint, Lap joint, T-
joint
Weld Position WPosition axis(x, y, z)
Weld Process Wprocess SMAW, SAW,
GMAW, GTAW
Gas type Gas type CO2 , argon, argon/co2
mix
Operation factor OF (SMAW, 30%),
(GMAW, 45%),
(MCAW, 55%),
(GMAW, FCAW, and
SAW processes, 100%)
Start point StartP (x,y,z)
End point EndP (x,y,z)
79
6.3 The data structure of welding cost feature
By using the data structure of welding, we can define welding cost feature class as:
- Cost of material (C mtl) = (amount of welding wire C Rod) + (amount of protection paint C
Protection) + (gas volume rate C Gas)
- Cost of labor (C lab) = (t) C labor Operating Factor
- Cost of machine (C mch) = (C occp T1) + (C setup T2) + (C clean T3) + (C paint T4)
- Cost of welding (C weld) = f (Type of welding, welding condition)
- Cost of overhead (C OH) = f(C administration, C depreciation, Infrastructure)
CF (): cost feature: CF = [ , , , , ]
Figure 6.5 Data structure of welding cost feature class
Data structure of welding cost
feature class
Properties:
- Geometry information of first
work piece: L1, W1, H1
- Geometry information of
second work piece: L2, W2, H2
- Welding times: T1,T2,T3, T4
- Technical parameters - Welding geometric parameters
……………
Constraints:
- = $24 per hr
- < T1 * 3$
……………
Functions:
- Get-parameters ();
- Calculate-cost of material ();
- Calculate-cost of labor ();
- Calculate-cost of machine ();
……………….
80
6.4 Mapping of welding manufacture feature and the associative cost feature
To illustrate the cost feature association method, a feature-mapping system was used.
Feature-mapping system can map the cost features with other types of features in CE via
a set of data structures, and this data was derived from a CAD module or an ERP system.
The mapping function defined as a: MF = [WF, CF, f] where: f (WF) = CF
WF means welding feature, CF is the cost feature, and f represents the function
mapping algorithm. In the matrix below, we present a feature-mapping schematic
between the welding and cost features.
Mapping matrix Feature array Cost feature
*
=
(6-1)
81
6.5 Prototype results and discussion
To test the advantages of the proposed hybrid feature-based CE, a typical problem was
formulated as discussed in chapter 5 and the computer program was implemented. Visual
Studio 2012 was used to implement the CE engine as a functional module acting as a
driver in the prototype program; and an SQL Server 2012 was used to store and manage
our features and ERP data.
To obtain preliminary results, we selected a sample set of welding-related operation
data with 500 welding parts, currently manufactured by McCoy FARR, ranging from
simplistic to complex. This test case is large enough to serve as a good benchmark for our
proposed formulation, and yet small enough to be tractable, thus allowing us to test our
hypothesis. The test case was selected to mimic real welded parts and a real welding
feature-based CE problem as close as possible to reality, ensuring the validity of the
predicted results.
In the first step, we extracted data from the company’s ERP system for data analysis.
Recognizing the trend data, reasonably established by common numerical analysis, is the
next step. Data trends can illustrate the overall pattern of data changes, which is suitable
for comparing different groups of data. To determine data trends, more details need to be
added, such as machining and overhead costs. Besides adding more details, the new
welding cost is calculated based on a regression cost formula, which was introduced in
section 5.6. The data analysis shows that the new cost increases 7%-9% compared to the
primary company’s data.
The regression welding CE is presented below in table 6.2, beside the primary welding
cost. As a result, accurate CE based on a forward approach is used to cross validate the
DM cost estimation process.
82
Table 6.2 Sample CE based on regression analysis
Part ID Joint
type
Labor
cost per
hour ($)
Welding
time
(hr.)
Operatin
g factor
(%)
Amount
of
welding
wire
(kg)
Per rod
cost ($)
Paint
cost ($)
Gas
volume
rate
( )
Gas cost
per cubic
meter ($)
Machine
cost ($)
overhead
cost ($)
Primary
weld
cost ($)
Regression
weld cost
($)
Predicted
cost ($)
A101
butt-
joint 24 0.16 0.4 0.03 1.44 0.25 0.62 10.5 0.48 0.65 8.43 9.438 9.427
A102
butt-
joint 24 0.25 0.4 0.075 1.44 0.25 0.62 10.5 0.75 0.8 9.65 10.859 11.17
A103
butt-
joint 24 0.33 0.4 0.062 1.44 0.25 0.62 10.5 0.99 0.9 10.5 11.85 11.849
A104
butt-
joint 24 0.5 0.4 0.75 1.44 0.25 0.62 10.5 1.5 0.98 12.62 14.156 14.155
A105 T-joint 24 0.16 0.4 0.06 1.44 0.25 0.62 10.5 0.48 0.66 8.44 9.449 9.449
A106
butt-
joint 24 0.33 0.4 0.03 1.44 0.25 0.62 10.5 0.99 0.75 10.47 11.742 11.742
A107
butt-
joint 24 0.16 0.4 0.25 1.44 0.25 0.62 10.5 0.48 0.82 11.1 12.763 12.52
A108
Lap-
joint 24 0.33 0.4 0.046 1.44 0.5 0.84 10.5 0.99 0.76 10.54 12.288 11.801
83
Based on existing information for welding costs and features, we need to design a set
of rules (tailored to a specific implementation) that allowed us to define the guidelines
used to implement the DM cost estimation technique consistently, for the case study. The
set of rules we defined is the following:
• Rule 1: Joint type: {butt-joint, Lap-joint, T-joint}
• Rule 2: Labor cost ($): {hourly = 24}
• Rule 3: Welding time: {setup time, length of weld, welding sections, complexity,
machining volume rate}
• Rule 4: Operating factors: {FCAW= 40%}
• Rule 5: Amount of welding wire: {welding volume, filling metal density}
• Rule 6: Rod cost($): {per rod 7018=1.44}
• Rule 7: Paint cost($): {welding sections, joint type (butt-joint = 0.25, Lap-joint =
0. 5, T-joint = 0.25)}
• Rule 8: Gas volume rate: {nozzle size ½ inch = 0.62 (m³/hr.), nozzle size 5/8 inch
= 0.84(m³/hr.)}
• Rule 9: Gas cost($): {per cubic meter= 10 }
• Rule 10: Machine cost($): {welding time *3}
• Rule 11: Overhead cost($): {0.05*direct cost}
The proposed feature-based cost engine uses the following steps to prepare DM
cost estimation:
1. Reorganizing and setting cost feature data extracted from the ERP system
2. Implement data mining clustering algorithm via WEKA: Expectation
Maximization (EM) selected as a data mining clustering algorithm, which models
a data set based on linear combination and normal distribution [84]. EM divided a
data set into seven clusters, and then each cluster was used as an input for CE.
Eighty percent of each cluster was used as training data set and 20% for testing
the data set.
3. Implement data mining algorithms via WEKA: Linear Regression (LR); Multiple
Regression (MR); Artificial Neural Network (ANN), k-Nearest Neighbors
84
(KNN), and the Meteorological parameter section (MP5). For instance, figure 6.6
illustrate the implementation of the MP5 algorithm in WEKA. This model,
generated for each cluster, was saved and reused for further CE.
Figure 6.6 Sample result of CE based on the regression analysis of weka
Five hundred welding parts were used as an input for five generated data mining
models. The comparison between the results of five different algorithms in the case study
is shown in table 6.3 Clearly, ANN and MP5 have better results for correct prediction. In
addition, ANN has a better confidence level compared to the other algorithms. To
optimize the result of models, the trends of MMRE were analyzed. LR can be optimized
by adding MMRE to the CE. However, other models do not have a same trend, so the
result is sometimes less than the actual cost and sometimes more. In this research, the
suggested algorithm for welding CE is the artificial neural networks (ANN) algorithm.
M5P algorithm
Different rules
that generated
by algorithm
for calculating
total cost
Total rules that generated
The model that
can be store
and reuse
85
Table 6.3 correct matching and error rates comparison between five different algorithms to the historical data set (%)
Models N LR MR ANN KNN MP5
Correct
prediction
training
data
testing
data
MMRE training
data
testing
data
MMRE training
data
testing
data
MMRE training
data
testing
data
MMRE training
data
testing
data
MMRE
Cluster 0 52 78% 53% 53.4% 74% 68% 33% 84% 92% 16.8% 72% 86% 51.6% 85% 95% 33%
Cluster 1 130 68% 43% 52.1% 69.3% 29.6% 52% 82.3% 96.2% 15.6% 69.3% 35% 65.6% 79.6% 92.3% 25.3%
Cluster 2 78 82% 85% 45% 85% 89.6% 16.2% 79.6% 93.6% 16.5% 86% 70.3% 39.6% 74.6% 89.6% 35.6%
Cluster 3 43 76% 69% 33% 63.3% 46.2% 39.3% 85.6% 85% 12.3% 64.3% 35.7% 52.3% 89.7% 92.6% 55.6%
Cluster 4 19 68% 73% 52.3% 69.6% 36.1% 23.6% 80.3% 86.3% 16.3% 82.6% 73.5% 33.5% 86% 92.3% 38.9%
Cluster 5 55 64% 35% 59.6% 67.4% 42.5% 17.6% 78.6% 92% 19.3% 68.9% 75% 59% 84% 85.6% 18.9%
Cluster 6 123 74% 46% 43.8% 72.6% 39.6% 56.4% 83.6% 93.5% 25.2% 74.8% 48.6% 45% 84.6% 91.6% 47.6%
AVG 73% 58% 48% 72% 50% 34% 82% 91% 17% 74% 61% 50% 83% 91% 36%
86
Figure 6.7 illustrates more extended welding cost information, which includes
following inputs:
-Joint type: butt-joint, Lap-joint, T-joint
-Welding length: the length of welding seam (mm)
-Welding time: setup time, paint time, clean time, weld pointing time, total welding time
(hr)
-Operating factor: (%)
-Labor cost: labor cost per hour, labor working cost ($)
-Paint cost: ($)
-Rod cost: amount of welding wire, rod diameter, per rod cost, total rod cost ($)
-Power cost: ($)
-Machine cost: ($)
-Overhead cost: ($)
-Gas cost: gas volume rate, amount of gas used, gas cost per cubic meter, type of gas, gas
cost by feature, total gas cost ($)
This kind of information should be provide by user, however if some data is not
available, the UI will be fill it based on historical data. This UI is intended to allow cost
engineers to review and capture all welding cost features; then the cost can be calculated
based on parametric CE and five data mining algorithms. Finally, the company can select
the best algorithm based on the company’s strategy for CE.
87
Figure 6.7 User Interface (UI) for welding cost estimation
Output
88
6.6 Conclusion
This research presents a hybrid CE methodology with feature-based empirical data
regression and data mining algorithms. The case study shows an application to welding
products, and it demonstrated that the approach is capable of producing satisfactory
results for different welding-feature-based parts. The proposed model has proven to be
repeatable, robust, and accurate for the range of features applied. In addition, as shown in
the case study, a set of rules or guidelines used to fill the gaps in missing data. Such rules
are complimentary with specific applications while the built-in features consider general
patterns of welding parameter estimation. This estimation methodology could be
extended beyond the cost of engineering only, e.g., the total manufacturing time
evaluation.
However, there are some limitations of which the reader should be aware:
- Predefinition of manufacturing features is the first limitation. The manufacture
process data can be consistently managed and analyzed. The consistent definitions
need one-time semantic modeling effort in any adopting enterprise. However,
reusing the definitions with repetitive application is highly recommended and
hence the implementation barrier would be reduced significantly.
- Data mining has to be supported with available and comprehensive historical data.
- The another limitation of this proposed method is that currently, in our case, ERP
data has to translated into a different data structure in order to support data mining
and feature data extraction. However, hopefully, with the popular adoption of the
proposed scheme and well-defined feature. Then, the ERP data structure can be
interfaced automatically with a converter or a compliance standard. Clearly, a lot
of future work is necessary.
- The proposed system is based on a semi-automatic hybrid method, where
categorizing the product features based on their shapes and characteristics has to
89
be done by engineers. Potentially, this task could be assisted by a feature
recognition program [85], which is not the focus of this study.
- The company’s ERP data description is not the same as our welding and cost
feature- definitions. We preprocessed the data based on the definitions proposed,
to keep the consistency of data input to the prototype algorithm. This issue arose
from the fact that in some instances, the actual manufacturing method applied is
not the same as the process plan, causing some discrepancy in our testing model.
- Finally, some features values do not exist in ERP system and we have to find
them based on other sources. For example, to find the thickness of welding
specifications of the seam, information has to be extracted from a CAD/CAM
model.
90
Chapter 7: Conclusion and future work
7.1 Conclusion
The main subject of this research was developing a concept of unique feature type for
cost estimation purpose. A new category of feature, the cost feature is defined as a class
in the unified feature modelling system to address the characteristics of cost engineering
entities, constraints and dependency relations.
Firstly, this study proposes the semantic model for automatic product cost estimation.
This model integrates three functional sub-models: feature-based costing, data mining,
and semantic reasoning. The goal of this model is investigate a concept of new
manufacturing cost calculation model coherently throughout the lifecycle of a product
series, especially emphasizing at the conceptual design stage.
Secondly, this study describes a new cost estimation functional module that can be
implemented in an ERP system including the representative of auxiliary data, machining
and design models. Some key concepts in the module, such as cost feature library, the
relation among sub-models and feature mapping are proposed.
Thirdly, this study proposes a hybrid cost estimation method by combining traditional
empirical and advanced data-mining techniques. The goal of this method is providing the
more accurate cost estimation via contrasting a user interface that can show the cost
estimation cross-checking.
The author believes that there are, significant advantages in implementing the
proposed hybrid methodology:
91
- Capability to work out the estimated confidence level from different components
- Less time and complexity in application compared to the traditional analytical
cost-estimation formula
- Reusability and recursive accuracy enhancement over cost patterns and
estimations from DM results based on accumulated historical data
- Support new part CE without detailed exhaustive data
- Feasibility of applying in a manufacturing company within a ERP system via
customized UIs for increased CE accuracy and short feedback time to the
decision-makers and customers
- A case that may benefit from the methods proposed in this case study
- The results of case study support this argument and provide additional guidelines
to cost engineers
7.2 Future work
This study has illustrates the benefits of using cost feature for dynamic, complex, and
bottom-up cost estimation. Furthermore, the results suggested the more accurate data
mining algorithm for welding cost estimation. However, it is to be further proven whether
other companies will experience similar improvements. To address the generality of
applications, future work must be done in the following fields:
- Develop a comprehensive cost estimation model for other types of operations
rather than welding
- Advanced cost analysis for products with two or more operations
- Apply method to other industries
92
References
[1] Kaplan, Robert S., ed. Measures for manufacturing excellence. Harvard Business
Press, 1990.
[2] Mileham, A. R., G. C. Currie, A. W. Miles, and D. T. Bradford. "A parametric
approach to cost estimating at the conceptual stage of design." Journal of engineering
design 4, no. 2 (1993): 117-125.
[3] Fritz, Ronald D., and D. L. Kimbler. "Traditional product cost estimation: problems
and alternative''." In Proceedings of the 5th Industrial Engineering Research Conference,
Norcross, GA, pp. 375-80. 1996.
[4] Robert Jacobs, F. "Enterprise resource planning (ERP)—A brief history."Journal of
Operations Management 25, no. 2 (2007): 357-363.
[5] Daneva, Maya, and Roel Wieringa. "Cost estimation for cross-organizational ERP
projects: research perspectives." Software Quality Journal 16, no. 3 (2008): 459-481.
[6] Hong, Kyung-Kwon, and Young-Gul Kim. "The critical success factors for ERP
implementation: an organizational fit perspective." Information & Management40, no. 1
(2002): 25-40.
[7] Herrmann, Jeffrey W. "Design for production: concepts and applications." In
Proceedings of the SME East Coast Region 3 Annual Members Conference, Bethlehem,
PA, USA, October 10-11. Hiller Aircraft Corporation, 2003.
[8] Woods, Donald R. Cost Estimation for the Process Industries. McMaster University,
1983.
[9] Trivailo, O., M. Sippel, and Y. A. Şekercioğlu. "Review of hardware cost estimation
methods, models and tools applied to early phases of space mission planning." Progress
in Aerospace Sciences 53 (2012): 1-17.
93
[10] Holmes, Geoffrey, Andrew Donkin, and Ian H. Witten. "Weka: A machine learning
workbench." Intelligent Information Systems, 1994. Proceedings of the 1994 Second
Australian and New Zealand Conference on. IEEE, 1994.
[11] Ma, Yongsheng. Semantic Modeling and Interoperability in Product and Process
Engineering. Springer London Limited, 2013.
[12] Niazi, Adnan, Jian S. Dai, Stavroula Balabani, and Lakmal Seneviratne. "Product
cost estimation: Technique classification and methodology review." Journal of
manufacturing science and engineering 128, no. 2 (2006): 563-575.
[13] Clark, Forrest, and Albert B. Lorenzoni. Applied cost engineering. CRC Press, 2011.
[14] Stewart, Rodney D., Richard M. Wyskida, and James D. Johannes, eds. Cost
estimator's reference manual. Vol. 15. John Wiley & Sons, 1995.
[15] Rush, Christopher, and Rajkumar Roy. "Analysis of cost estimating processes used
within a concurrent engineering environment throughout a product life cycle." 7th ISPE
International Conference on Concurrent Engineering: Research and Applications, Lyon,
France, July 17th-20th, Technomic Inc., Pennsylvania USA. 2000.
[16] Ansari, Shahid L., et al. Target costing. Irwin/McGraw-Hill, 1997.
[17] Roy, Rajkumar. "Cost engineering: why, what and how?." (2003).
[18] Ostwald, Phillip F., and Timothy S. McLaren. Cost analysis and estimating for
engineering and management. Pearson Education, 2004.
[19] Sweeting, Jack. Project Cost Estimating: Principles and Practice. IChemE, 1997.
[20] Shuford Jr, R. H. "Activity-based costing and traditional cost allocation
structures." Cost estimator’s reference manual, 2nd ed., New York, Johan Wiley & Sons
Inc (1995): 41-94.
94
[21] Brinke, Erik. Costing support and cost control in manufacturing. A cost estimation
tool applied in the sheet metal domain. University of Twente, 2002.
[22] Asiedu, Y., and P. Gu. "Product life cycle cost analysis: state of the art
review." International journal of production research 36.4 (1998): 883-908.
[23] Hart, C. G., et al. "An advanced cost estimation methodology for engineering
systems." Systems Engineering 15.1 (2012): 28-40.
[24] Roy, Rajkumar, et al. "Quantitative and qualitative cost estimating for engineering
design." Journal of Engineering Design 12.2 (2001): 147-162.
[25] Layer, Alexander, et al. "Recent and future trends in cost estimation."International
Journal of Computer Integrated Manufacturing 15.6 (2002): 499-510.
[26] Datta, Partha P., and Rajkumar Roy. "Cost modelling techniques for availability type
service support contracts: a literature review and empirical study." CIRP Journal of
Manufacturing Science and Technology 3.2 (2010): 142-157.
[27] Rehman, Sumaira, and Marin D. Guenov. "A methodology for modelling
manufacturing costs at conceptual design." Computers & industrial engineering35.3
(1998): 623-626.
[28] Ficko, M., et al. "Prediction of total manufacturing costs for stamping tool on the
basis of CAD-model of finished product." Journal of Materials Processing
Technology 164 (2005): 1327-1335.
[29] Sol, Henk G., A. Th Cees, and Pieter F. de Vries Robbé, eds. Expert systems and
artificial intelligence in decision support systems. Springer, 1987.
[30] Kingsman, Brian G., and Antonio Artur de Souza. "A knowledge-based decision
support system for cost estimation and pricing decisions in versatile manufacturing
companies." International Journal of Production Economics 53.2 (1997): 119-139.
95
[31] Partridge, Derek, and Khateeb M. Hussain. "Knowledge Based Information
Systems." Mcgraw-Hill Book Comp., 1995.
[32] Gayretli, A., and H. S. Abdalla. "An object-oriented constraints-based system for
concurrent product development." Robotics and Computer-Integrated Manufacturing 15.2
(1999): 133-144.
[33] Nikolopoulos, Chris. Expert systems: introduction to first and second generation and
hybrid knowledge based systems. Marcel Dekker, Inc., 1997.
[34] Smith, Alice E., and Anthony K. Mason. "Cost estimation predictive modeling:
Regression versus neural network." The Engineering Economist 42.2 (1997): 137-161.
[35] Martin, Patrick, J-Y. Dantan, and Ali Siadat. "Cost estimation and conceptual
process planning." Digital Enterprise Technology. Springer US, 2007. 243-250.
[36] Asking, Pontus, and Stefan Gustavsson. "Cost breakdown analysis-A study of the
product costs in the kitchen appliances at IKEA of Sweden." (2011).
[37] Asking, Pontus, and Stefan Gustavsson. "Cost breakdown analysis-A study of the
product costs in the kitchen appliances at IKEA of Sweden." (2011).
[38] Shah, Jami J., and Mary T. Rogers. "Functional requirements and conceptual design
of the feature-based modelling system." Computer-Aided Engineering Journal 5.1 (1988):
9-15.
[39] Emmerik, Van, M.J.G.M. and Jansen, F.W. “User interface for feature modelling, in
Proceedings” of CAPE '89, Tokyo, October 1989 (F.K. KIMURA & A. ROLSATDAS
(Eds), pp. 625-632).
[40] Wierda, Leo S. "Design-oriented cost information: the need and the
possibilities." Journal of Engineering Design 1.2 (1990): 147-167.
96
[41] Wierda, Leo S. "Linking design, process planning and cost information by feature-
based modelling." Journal of Engineering Design 2.1 (1991): 3-19.
[42] Zhang, Y. F., Jerry YH Fuh, and W. T. Chan. "Feature-based cost estimation for
packaging products using neural networks." Computers in Industry 32.1 (1996): 95-113.
[43] Ou-Yang, C., and T. S. Lin. "Developing an integrated framework for feature-based
early manufacturing cost estimation." The international journal of advanced
manufacturing technology 13.9 (1997): 618-629.
[44] Leibl, Peter. "Cost calculation with a feature-based CAD system using modules for
calculation, comparison and forecast." Journal of Engineering Design 10.1 (1999): 93-
102.
[45] Brimson, J. A. "Feature costing: beyond ABC." Journal of Cost Management 12
(1998): 6-13.
[46] Masmoudi, Faouzi, Wafik Hachicha, and Zoubeir Bouaziz. "A new feature-concept
applied in cost estimation model for a weld assemblage: Additional Information." (2007):
149-162.
[47] Masmoudi, Faouzi, Zobeir Bouaziz, and Wafik Hachicha. "Computer-aided cost
estimation of weld operations." The International Journal of Advanced Manufacturing
Technology 33.3-4 (2007): 298-307.
[48] Eklin, Mark, Yohanan Arzi, and Avraham Shtub. "Model for cost estimation in a
finite-capacity stochastic environment based on shop floor optimization combined with
simulation." European journal of operational research 194.1 (2009): 294-306.
[49] Feldman, P., and A. Shtub. "Model for cost estimation in a finite-capacity
environment." International journal of production research 44.2 (2006): 305-327.
97
[50] Wei, Jingxing, and Y-S. Ma. "Design of a feature-based order acceptance and
scheduling module in an ERP system." Computers in Industry 65.1 (2014): 64-78.
[51] Khalifelu, Zeynab Abbasi, and Farhad Soleimanian Gharehchopogh. "Comparison
and evaluation of data mining techniques with algorithmic models in software cost
estimation." Procedia Technology 1 (2012): 65-71.
[52] Lum, Karen T., Daniel R. Baker, and Jairus M. Hihn. "The effects of data mining
techniques on software cost estimation." Engineering Management Conference, 2008.
IEMC Europe 2008. IEEE International. IEEE, 2008.
[53] Jørgensen, Magne. "A critique of how we measure and interpret the accuracy of
software development effort estimation." First International Workshop on Software
Productivity Analysis and Cost Estimation. Information Processing Society of Japan,
Nagoya. 2007.
[54] Nguyen, Vu, Bert Steece, and Barry Boehm. "A constrained regression technique for
COCOMO calibration." Proceedings of the Second ACM-IEEE international symposium
on Empirical software engineering and measurement. ACM, 2008.
[55] Hughes, Robert T. "Expert judgement as an estimating method." Information and
software technology 38.2 (1996): 67-75.
[56] Gordon, Leonard. "Using Classification and Regression Trees (CART) in SAS®
Enterprise Miner TM For Applications in Public Health." Data Mining and Text
Analytics (2013): 089-2013
[57] Bidarra, R. A. F. A. E. L., and Willem F. Bronsvoort. "Semantic feature modelling."
Computer-Aided Design 32, no. 3 (2000): 201-225.
[58] Ma, Yongsheng, N. Sajadfar, L. Campos Triana "A feature-based semantic model
for automatic product cost estimation" IACSIT International Journal of Engineering and
Technology, 6(2), 2014, 109-113.
98
[59] Ma, Yongsheng, T. Tong "Associative feature modelling for concurrent engineering
integration" Computers in Industry 51, no. 1 (2003): 51-71.
[60] Al Khalidi, Nesreen, Ahmad A. Saifan, and Izzat M. Alsmadi. "Selecting a standard
set of attributes for cost estimation of software projects." In Computer, Information and
Telecommunication Systems (CITS), 2012 International Conference on, pp. 1-5. IEEE,
2012.
[61] Chan, Felix TS, and Jie Zhang. "Modelling for agile manufacturing systems."
International Journal of Production Research 39, no. 11 (2001): 2313-2332.
[62] Svendsen, Gunnar Lind Haase, and Jens Fyhn Lykke Sørensen. "There's more to the
picture than meets the eye: Measuring tangible and intangible capital in two marginal
communities in rural Denmark." Journal of Rural Studies 23, no. 4 (2007): 453-471.
[63] Han, JungHyun, Mike Pratt, and William C. Regli. "Manufacturing feature
recognition from solid models: a status report." Robotics and Automation, IEEE
Transactions on 16, no. 6 (2000): 782-796.
[64] Lin, Than, Jae-Woo Lee, and E. L. J. Bohez. "New integrated model to estimate the
manufacturing cost and production system performance at the conceptual design stage of
helicopter blade assembly." International Journal of Production Research 50, no. 24
(2012): 7210-7228.
[65] Ulrich, Karl T., and Steven D. Eppinger. Product design and development. Vol. 384.
New York: McGraw-Hill, 1995.
[66] Liu, Zhenyu, Wanghui Bu, and Jianrong Tan. "Motion navigation for arc welding
robots based on feature mapping in a simulation environment." Robotics and Computer-
Integrated Manufacturing 26, no. 2 (2010): 137-144.
99
[67] Zhang, De-Long, Jiaqing Qiao, Jun-Bao Li, Li-Yan Qiao, Shu-Chuan Chu, and John
F. Roddick. "Optimizing Matrix Mapping with Data Dependent Kernel for Image
Classification." (2014).
[68] Bernstein, Abraham, Foster Provost, and Shawndra Hill. "Toward intelligent
assistance for a data mining process: An ontology-based approach for cost-sensitive
classification." Knowledge and Data Engineering, IEEE Transactions on 17, no. 4 (2005):
503-518.
[69] Shaw, N. K., M. Susan Bloor, and A. De Pennington. "Product data models."
Research in Engineering Design 1, no. 1 (1989): 43-50.
[70] Peckham, Joan, and Fred Maryanski. "Semantic data models." ACM Computing
Surveys (CSUR) 20, no. 3 (1988): 153-189.
[71] Zave, José Nuno Oliveira Pamela. "FME 2001: Formal Methods for Increasing
Software Productivity."
[72] Shaul, Levi, and Doron Tauber. "CSFs along ERP life-cycle in SMEs: a field
study." Industrial Management & Data Systems 112.3 (2012): 360-384.
[73] Li, Conghua. "ERP packages: what's next?." Information Systems Management16
(1999): 31-35.
[74] Veeramani, Dharmaraj, and Pawan Joshi. "Methodologies for rapid and effective
response to requests for quotation (RFQs)." IIE transactions 29.10 (1997): 825-838.
[75] “Epicor Enterprise Resource Planning Overview”,
http://www.atc.gr/files/Brochures/Epicor-Enterprise-Resource-Planning-Overview-
ENS.PDF
[76] http://www.infor.com/product_summary/erp/visual/
100
[77] Sajadfar, Narges, Luis Campos Triana and Ma Y.-S. "Interdisciplinary semantic
interactions within a unified feature model for product cost estimation." International
Journal of Mechanical Engineering and Mechatronics (2014): 10-19
[78] Cao, Lian Min, Xi Wei Wang, and Qing Liang Zeng. "Study on Cost Modeling
Based on Feature Mapping Technology." Advanced Materials Research 295 (2011): 830-
833.
[79] Hall, Mark, et al. "The WEKA data mining software: an update." ACM SIGKDD
explorations newsletter 11.1 (2009): 10-18.
[80] Gunduz, Murat, Latif Onur Ugur, and Erhan Ozturk. "Parametric cost estimation
system for light rail transit and metro trackworks." Expert Systems with
Applications 38.3 (2011): 2873-2877.
[81] Seo, Kwang-Kyu. "A methodology for estimating the product life cycle cost using a
hybrid GA and ANN model." Artificial Neural Networks–ICANN 2006. Springer Berlin
Heidelberg (2006): 386-395.
[82] Gharehchopogh, Farhad. "Neural networks application in software cost estimation:
A case study"
International Symposium on Innovations in Intelligent Systems and Applications
(INISTA 2011): 69-73.
[83] Wu, Xindong, and Vipin Kumar, eds. "The top ten algorithms in data mining". CRC
Press, 2009
[84] Abbas, Osama Abu. "Comparisons Between Data Clustering
Algorithms."International Arab Journal of Information Technology (IAJIT) 5.3 (2008).
101
[85] Prabhakar, Shashikanth, and Mark R. Henderson. "Automatic form-feature
recognition using neural-network-based techniques on boundary representations of solid
models." Computer-Aided Design 24.7 (1992): 381-393.
[86] McCoy Drilling and Completions (Farr) Company, product drawings, internal
document, 2013.
